Monday, September 4, 2023

Measuring signal dynamics of the RX-888 (Mk2)

As a sort of follow-up to the previous posting about the RX-888 (Mk2) I decided to make some measurements to help characterize the gain and attenuation settings.

The RX-888 (Mk2) has two mechanisms for adjusting gain and attenuation:

  • The PE4312 attenuator.  This is (more or less) right at the HF antenna input and it can be adjusted to provide up to 31.5dB of attenuation in 0.5dB steps.
  • The AD8370 PGA.  This PGA (Programmable Gain Amplifier) can be adjusted to provide a "gain" from -11dB to about 34dB.


While this blog posting has specific numbers related to the RX-888 (Mk2), its general principles apply to ALL receivers - particularly those operating as "Direct Sampling" HF receivers.  A few examples of other receivers in this category include the KiwiSDR and Red Pitaya - to name but two.

Other article RX-888 article:

I recently posted another article about the RX-888 (Mk2) discussing the thermal properties of its mechanical construction - and ways to improve it to maximize reliability and durability.  You can find that article here:  Improving the thermal management of the RX-888 (Mk2) - link.

* * * * *

Taking measurements

To ascertain the signal path properties of an RX-888 (Mk2) I set its sample rate to 64 Msps and using both the "HDSDR" and "SDR Radio" programs (under Windows - because it was convenient) and a a known-accurate signal generator (Schlumberger Si4031) I made measurements at 17 MHz which follow:

Gain setting (dB)Noise floor (dBm/Hz)Noise floor (dBm in 500Hz)Apparent Clipping level (dBm)

Figure 1:  Measured performance of an RX-888 Mk2.  Gain mode is "high" with 0dB attenuation selected.

For convenience, the noise floor is shown both in "dBm/Hz" and in dBm in a 500 Hz bandwidth - which matches the scaling used in the chart below.  As the programs that I used have no direct indication of A/D converter clipping, I determined the "apparent" clipping level by noting the amplitude at which one additional dB of input power caused the sudden appearance of spurious signals.  Spot-checking indicated that the measured values at 17 and 30 MHz were within 1 dB of each other on the unit being tested.

Determining the right amount of "gain"

It should be stated at the outset that most of the available range of gain and attenuation provided by the RX-888's PE4312 step attenuator and AD8370 variable gain amplifier are completely useless to us.  To illustrate this point, let's consider a few examples.

Consider the chart below:

Figure 2:  ITU chart showing various noise environments versus frequency.

This chart - from the ITU - shows predicted noise floor levels - in a 500 Hz bandwidth - that may be expected at different frequencies in different locations.  Anecdotally, it is likely that in these days of proliferating switch-mode power supplies that we really need another line drawn above the top "Residential" curve, but let's be a bit optimistic and presume that it still holds true these days.

Let us consider the first entry in Figure 1 showing the gain setting of 0dB.  If we look at the "Residental" chart, above, we see that the curve at 30 MHz indicates a value very close to the -113dBm value in the "dBm in 500 Hz" column.  This tells us several things:

  • Marginal sensitivity.  Because the noise floor of the RX-888 (Mk2) and that of our hypothetical RF environment are very close to each other, we may not be able to "hear" our noise floor at 30 MHz (e.g. the 10 meter amateur band).  One would need to do an "antenna versus no antenna" check of the S-meter/receiver to determine if the former causes an increase in signal level:  If not, additional gain may be needed to be able to hear signals that are at the noise floor.
  • More gain may not help.  If we do perform the "antenna versus no antenna" test and see that with the antenna connected we get, say, an extra S-unit (6dB) of noise, we can conclude that under those conditions that more gain will not help in absolute system sensitivity.

Thinking about the above two statements a bit more, we can infer several important points about operating this or any receiver in a given receive environment:

  • If we can already "hear" the noise floor, more gain won't help.  In this situation, adding more gain would be akin to listening to a weak and noisy signal and expecting that increasing the volume would cause the signal to get louder - but not the noise.  
  • More gain than necessary will reduce the ability of the receiver to handle strong signals.  The HF environment is prone to wild fluctuations and signals can go between well below the local noise floor and very strong, so having any more gain that you need to hear your local noise floor is simply wasteful of the receiver's signal handling capability.  This fact is arguably more important with wide-band, direct-sampling receivers where the entire HF spectrum impinges on the analog-to-digital converter rather than a narrow section of a specific amateur band as is the case in "conventional" analog receivers.

Let us now consider what might happen if we were to place the same receiver in an ideal, quiet location - in this case, let's look at the "quiet rural" (bottom line) on the chart in Figure 2.

Again looking at the value at 30 MHz, we see that our line is now at about -133dBm (in 500 Hz) - but if we have our RX-888 gain set at 0 dB, we are now ((-133) - (-113) = ) 20 dB below the noise floor.  What this means is that a weak signal - just at the noise floor - is more than 3 S-units below the receiver sensitivity.  This also means that a receiver that may have been considered to be "Okay" in a noisy, urban environment will be quite "deaf" if it is relocated to a quiet one.

In this case we might think that we would simply increase our gain from 0 dB to +33dB - but you'll notice that even at that setting, the sensitivity will be only -131dBm in 500 Hz - still a few dB short of being able to hear the noise in our "antenna versus no antenna" test.

Too much gain is worse than too little!

At this point I refer to the far-right column in Figure 1 that shows the clipping level:  With a gain setting of +33dBm, we see that the RX-888 (Mk2) will overload at a signal level of around -31dBm - which translates to a  signal with a strength a bit higher than "S9 + 40dB".  While this sound like a strong signal, remember that this signal level is the cumulative TOTAL of ALL signals that enter the antenna port.  Thinking of it another way, this is the same as ten "S9+30dB" signals or one hundred "S9+20dB" signals - and when the bands are "open," there will be many times when this "-31dBm" signal level is exceeded from strong shortwave broadcast signals and lightning static.

In the case of too-little gain, only the weakest signals, below the receiver's noise floor will be affected - but if the A/D converter in the receiver is overloaded, ALL signals - weak or strong - are potentially disrupted as the converter no longer provides a faithful representation of the applied signal.  When the overload source is one or more strong transmissions, a melange of all signals present is smeared throughout the receive spectrum consisting of many mixing products, but if the overload is a static crash, the entire receive spectrum can be blanked out in a burst of noise - even at frequencies well removed from the original source of static.

Most of the adjustment range is useless!

Looking carefully at Figure 1 at the "noise floor" columns, you may notice something else:  Going from a gain of 0 dB to 10 dB, the noise floor "improves" (is lower) by about the same amount - but if you go from 25 dB gain to 33 dB gain we see that our noise floor improves by only 1 dB - but our overload threshold changes by the same eight dB as our gain increase.

What we can determine from this is that for practical purposes, any gain setting above 20 dB will result in a very little receiver sensitivity improvement while causing a dramatic reducing in the ability of the receiver to handle strong signals.

Based on our earlier analysis in a noise "Urban" environment, we can also determine that a gain setting lower than 0 dB will also make our receiver too-insensitive to hear the weakest signals:  The gain setting of -25dB shown in Figure 1 with a receive noise floor of -79dBm (500 Hz) - which is about S8 - is an extreme example of this.

Up to this point we have not paid any attention to the PE4312 attenuator as all measurements were taken with this set to minimum.  The reason for this is quite simple:  The noise figure (which translates to the absolute sensitivity of a receiver system) is determined by the noise generation of all of the components.  As reason dictates, if you have some gain in the signal path, the noise contribution of the devices after the gain have lesser effects - but any loss or noise contribution prior to the gain will directly increase the noise figure.


For examples of typical HF noise figure values, see the following articles:

Based on the articles referenced above, having a receiver system with a noise figure of around 15dB is the maximum that will likely permit reception at the noise floor of a quiet 10 meter location.  If you aren't familiar with the effects of noise figure - and loss - in a receive signal path, it's worth playing with a tool like the Pasternack Enterprises Cascaded Noise Figure Calculator (link) to get a "feel" of the effects.

I do not have the ability to measure the precise noise figure of the RX-888 (Mk2) - and if I did do so, I would have to make such a measurement using the same variety of configurations depicted in Figure 1 - but we can know some parameters about the worst-case:

  • Bias-Tee:  Estimated insertion loss of 1dB
  • PE4312:  Insertion loss of 1.5dB at minimum attenuation
  • RF Switch (HF/VHF) 1dB loss
  • 50-200 Ohm transformer:  1dB loss
  • AD8370 Noise figure:  8dB (at gain of 20dB)

The above sets the minimum HF floor noise figure of the RX-888 (Mk2) at about 12.5dB with an AD8370 gain setting of 20dB - but this does not include the noise figure of the A/D converter itself - which would be difficult to measure using conventional means.

On important aspect about system noise figure is that once you have loss in a system, you cannot recover sensitivity - no matter how much gain or how quiet your amplifier may be!  For example, if you have a "perfect" 20 dB gain amplifier with zero noise, if you place a 10 dB attenuator in front of it, you have just turned it into an amplifier with 10 dB noise figure with 10dB gain and there is nothing that can be done to improve it - other than get rid of the loss in front of the amplifier.

Similarly, if we take the same "perfect" amplifier - with 20dB of gain - and then cascade it with a receiver with a 20dB noise figure, the calculator linked above tells us that we now have a system noise figure of 3 dB since even with 20dB preceeding it, our receiver still contributes noise!

If we presume that the LTC2208 A/D converter in the RX-888 has a noise figure of 40dB and no gain (a "ballpark" value assuming an LSB of 10 microvolts - a value that probably doesn't reflect reality) our receive system will therefore have a noise figure of about 22dB.

What this means is that in most of the ways that matter, the PE4312 attenuator is not really very useful when the RX-888 (Mk2) is being used for reception of signal across the HF spectrum, in a relatively quiet location on an antenna system with no additional gain.

Where is the attenuator useful?

From the above, you might be asking under what conditions would the built-in PE4312 attenuator actually be useful?  There are two instances where this may be the case - and this would be applied ONLY if you have been unable to resolve overload situations by setting the gain of the AD8370 lower.

  • In a receive signal path with a LOT of amplification.  If your receive signal path has - say - 30dB of amplification (and if it does, you might ask yourself "why?") a moderate amount of attenuation might be helpful.
  • In a situation where there are some extremely strong signals present.  If you are near a shortwave or mediumwave (AM broadcast) transmitter that induces extremely strong signals in the receiver that cause intractable overload, the temporary use of attenuation may prevent the receiver from becoming overloaded to the point of being useless - but such attenuation will likely cause the complete loss of weaker signals.  In such a situation, the use of directional antennas and/or frequency-specific filtering should be strongly considered!

Improving sensitivity

Returning to an earlier example - our "Quiet Rural" receive site - we observed that even with the gain setting of the RX-888 (Mk2) at maximum, we would still not be able to hear our local noise floor at 30 MHz - so what can be done about this?

Let us build on what we have already determined:

  • While sensitivities is slightly improved with higher gain values, setting the gain above 20dB offers little benefit while increasing the likelihood of overload.
  • In a "Quiet Rural" situation, our 30 MHz noise floor is about -133dBm (500 Hz BW) which means that our receiver needs to attain a lower noise floor than this:  Let's presume that -136dBm (a value that is likely marginal) is a reasonable compromise.

With a "gain" setting of 20dB we know that our noise floor will be around -128dBm (500 Hz) and we need to improve this by about 8 dB.  For straw-man purposes, let's presume that the RX-888 (Mk2) at a gain setting of 20dB has a noise figure of 25dB, so let's see what it takes for an amplifier that precedes the RX-888 (Mk2) to lower than to 17dB or so using the Pasternak calculator above:

  • 10dB LNA with 7 dB noise figure:  This would result in a system noise figure of about 16 dB - which should do the trick.

Again, the above presumes that there is NO  loss (cable, splitters, filtering) preceding the preamplifier.  Again, the presumed noise figure of 25dB for the RX-888 (Mk2) at a gain setting of 20 is a bit of a "SWAG"  - but it illustrates the issue.

Adding a low-noise external amplifier also has another side-effect:  By itself, with a gain setting of +33, the RX-888 (Mk2)'s overload point is -31dBm, but if we reduce the gain of the RX-888 to 20dB the overload drops to -18dBm - but adding the external 10dB gain amplifier will effectively reduce the overload to -28dBm, but this is still 5 dB better than if we had turned the RX-888's gain all of the way up!

Taking this a bit further, let's presume that we use, instead, an amplifier with 3dB noise figure and 8 dB gain:  Our system noise figure is now about 17dB, but our overload point is now -26dBm - even better!

The RX-888 is connected to a (noisy) computer!

Adding appropriate amounts of external gain has an additional effect:  The RX-888 (and all other SDRs) are computer/network connected devices with the potential of ingress of stray signals from connected devices (computers, network switches, power supplies, etc.).  The use of external amplifiers can help override (and submerge) such signals and if proper care is taken to choose the amount of gain of the external amplification and properly choose gain/attenuation settings within the receiver, superior performance in terms of sensitivity and signal-handling capability can be the result.

Additional filtering

Only mentioned in passing, running a wideband, direct-sampling receiver of ANY type (be it RX-888, KiwiSDR, Red Pitaya, etc.) connected to an antenna is asking a lot of even 16 bits of conversion!  If you happen to be in a rather noisy, urban location, the situation is a bit better in the sense that you can reduce receiver gain and still hear "everything there is to hear" - but if you have a very quiet location that requires extra gain, the same, strong signals that you were hearing in the noisy environment are just as strong in the quiet environment.

Here are a few suggestions for maximizing performance under the widest variety of situations:

  • Add filtering for ranges that you do not plan to cover.  In most cases, AM band (mediumwave) coverage is not needed and may be filtered out.  Similarly, it is prudent to remove signals above that in which you are interested.  For the RX-888 (Mk2), if you run its sampling rate at just 65 MHz or so, you should install a 30 MHz low-pass filter to keep VHF and FM broadcast signals out.
  • Add "window" filtering for bands of interest.  If you are interested only in amateur radio bands, there are a lot of very strong signals outside the bands of interest that will contribute to overload of the A/D converter.  It is possible to construct a set of filters that will pass only the bands of interest - but this does not (yet?) seem to be a commercial product.  (Such a product may be available in the near future - keep a lookout here for updates.)
  • Add a "shelving" filter.  If you examine the graph in Figure 2 you will notice that as you go lower in frequency, the noise floor goes UP What this means is that at lower frequencies, you need less receiver sensitivity to hear the signals that are present - and it also means that if you increasingly attenuate those lower frequencies, you can remove a significant amount of RF energy from your receiver without actually reducing the absolute sensitivity.  A device that does just this is described in a previous blog article "Revisiting the limited-attenuation high-pass filter - again (link)".  While I do not offer such a filter personally, such a device - along with an integrated 30 MHz low-pass filter - may be found at Turn Island Systems - HERE.


  • The best HF weak-signal performance for the RX-888 (Mk2) will occur with the receiver configured for "High" gain mode, 0 dB attenuation and a gain setting of about 20dB.  Having said this, you should always to the "antenna versus no antenna" test:  If you see more than 6-10dB increase in the noise level at the quietest frequency, you probably have too much gain.  Conversely, if you don't see/hear a difference, you probably need more gain - taking care in doing so.
  • For best HF performance of this - or any other wideband, direct-sampling HF SDR (RX-888, KiwiSDR, Red Pitaya, etc.) additional filtering is suggested - particularly the "shelving" filter described above.
  • In situations where the noise floor is very low (e.g. a nice, receive quiet location) many direct-sampling SDRs (RX-888, KiwiSDR, Red Pitaya) will likely need additional gain to "hear" the weaker signals - particularly on the higher HF bands.  While some of these receivers offer onboard gain adjustment, the use of external high-performance (low-noise) amplification (along with filtering and careful adjustment of the devices' gain adjustments) will give improved absolute sensitivity while helping to preserve large-signal handling capability.
  • Because the RX-888 is a computer-connected device, there will be ingress of undesired signals from the computer and the '888's built-in circuitry.  The use of external amplification - along with appropriate decoupling (e.g. common-mode chokes on the USB cable and connecting coaxial cables) can minimize the appearance of these signals.


This page was stolen from



Tuesday, August 29, 2023

Improving the thermal management of the RX-888 (Mk2)

Figure 1:
The RX888 showing the "top" and RF connectors.  While
the heat sinks attached to the sides are visible, the large one
on the "bottom" plate are not.
Click on the image for a larger version.
The RX-888 Mk2 SDR is a USB3-based software-defined receiver that, unlike many others, is JUST and analog-to-digital converter (with a bit a low-pass filtering and adjustable attenuation and amplification) coupled to a USB 3 PHY chip.  With a programmable sample rate and a 65-ish MHz low-pass filter, it is capable of simultaneously inhaling the entire spectrum from a few 10s of kHz to about 60 MHz when run with a sample rate of 130 Msps - a rate which pretty much "maxes out" the USB 3 interface.

(Note:  There is also a frequency converter on board which will take up to a 10 MHz swath of spectrum between about 30 and 1800 MHz and shift it to a lower frequency within range of the A/D converter - but that's not part of this discussions.)

The purpose of this post is to discuss the thermal management of the RX-888 Mk2 which, in two words, can be described as  "marginal" and "inconsistent".

Other RX-888 article:

After posting this entry I produced another article about understanding the gain and properties of the HF signal path on the RX-888 (Mk2) - including information that can also be applied to other direct-sampling "all band HF" Software Defined Radios like the KiwiSDR, Red Pitaya and others.  You may read that article here:  Measuring signal dynamics of the RX-888 (Mk2) - Link.

Please note:

Despite the impression that the reader might get about the RX-888 (Mk2)'s thermal design and potential reliability, I would still consider it to be an excellent device at a good price - warts and all.

Its performance is quite good and especially since it lacks the FPGA that many other direct-sampling SDRs use, it is quite "future proof" in the sense that support of this receiver - and others like it that will no doubt appear soon - will be based on code running on the host computer (typically a PC or SBC) rather than on an FPGA contained within that requires specialized tools and knowledge for development and is limited by its own capacity.

If you think that an FPGA is needed, consider this:  For a few "virtual" receivers using "conventional" DSP techniques (e.g. HDSDR, SDR-Radio, etc.) a moderate Intel i7 is sufficient:  If using an optimized signal processing program like ka9q-radio along with a modest Intel i5, hundreds of virtual receivers covering the entire HF spectrum can be managed - but these are topics for another discussion.

In other words:  If you need a fairly simple, modestly-priced device to receive multiple RF channels it is well worth getting an RX-888 (Mk2) and performing some simple modification to it to improve its durability.  We can hope that future versions of this - and similar devices - will take these observations into account and produce even better hardware.

What's the problem?

There are scattered anecdotal reports of RX-888 (both the original and Mk2) simply "dying" after some period of time.  For most of these reports there are few details other than comments to this effect in various forums (e.g. little detailed analysis) but this was apparently enough of a problem with the original version of the RX-888 that with the Mk2, "improved" thermal management is one of the new features noted by its sellers.  (I do not have an original RX-888, but I would expect that the same general techniques could be applied to it as well.)

In short, here are a few comments regarding the thermal management of the RX-888 Mk2:

  • DO NOT run it outside its case.  There is a compressible thermal pad that goes between the exposed metal pad below the A/D converter that is intended to transfer heat to the case and without this in place the A/D converter and surrounding components can exceed 100C at moderate ambient temperatures.  If you plan to shuck the case, you should be aware of this and make appropriate arrangements to draw away heat via the same method. 

Figure 2:
Showing the paper double-sided "sticky tape" used to mount
the heat sinks.  Despite improper materials, these work "less
badly" than expected, but it's best to re-attach them properly.
Click on the image for a larger version.

  • The heat sinks are held on by double-sided tape.  The heat sink on the A/D converter appears to be some sort of thermal table like that seen on Raspberry Pi heat sink kits, but  those on the exterior of the case (one on each side, another the top) are held on with standard, paper-based double-sided tape:  People have reported these falling off with handling.  Additionally, because both the case and heat sinks are extruded their surfaces are not flat and all of the RX-888 (Mk2) units that I had a gap between the heat sink and the case through which a sheet of paper can be slid meaning that the heat sinks should be flattened a bit and/or attached using a material that will work as a thermally-conductive void filler.
  • The thermal pad may not be adequate.  Unless the small-ish thermal pad is placed precisely in its correct location, it will not be effective in its thermal transfer.  Additionally, these pads require a bit of compression between the board and the heat sink to be effective and it seems that the spacing between the board and the case is somewhat "loose" in the slot into which the PCB slides and that thermal contact may be inconsistent - more on this shortly.
  • Other components get very hot.  Next to the A/D converter are the 3.3 and 1.8 volt linear regulators which run very hot.  While this may be OK, they are next to (what appear to be) electrolytic capacitors which - if run very warm - can have rather short lifetimes.  While it is unknown if this is the case here, many regulators will become unstable (oscillate) if their associated capacitors degrade with lower capacitance and/or increased ESR (Equivalent Series Resistance) and if oscillation occurs due to capacitor degradation, this is likely to make the device unusable until the components are replaced.

Figure 3:
The top of the RX888 board.  The ADC's heat sink was
removed for the photo, but glued in place later to improve
its thermal transfer.
Click on the image for a larger version.

  • The FX3 USB interface chip can get very warm.  This chips is right next to the A/D converter.  There are anecdotal reports (again, nothing confirmed) that this particular chip can suffer reliability problems when running near its maximum rated temperature:  Whether this is due to a failure of silicon or (more likely) a mechanical failure of a solder connection on its BGA (ball grid array) as a result for thermal cycling remains to be seen, but either one could explain one of the RX-888's reported failure modes of no longer appearing to be the expected type of USB device, making the unit non-functional even though it seems to enumerate - albeit improperly.

Several different people have made spot measurements of the temperatures within an RX-888 and come up with different results, further indicating inconsistency in the efficacy of the passive cooling and showing the inherent difficulty in making such measurements - but here are a few comments that are likely relevant:

  • Unless you need coverage >30 MHz, do not run a sample rate higher than 65-70 Msps.  As with most devices, more current (and higher heat dissipation) will occur at a higher sample rate so keeping it well below its maximum (around 130 Msps) will reduce heating and potentially improve the lifetime.  
If you do run at a sample rate 64-70 Msps, it is recommended that a 30 MHz low-pass filter be installed as this will prevent aliasing due to this lower rate and the fact that the RX-888 (Mk2) has only a 60 MHz low-pass filter internally.
  • At normal "room" temperatures (68F/20C) the thermal properties of the RX-888 Mk2 are likely "Okay" if run at just 65-70 Msps - but increasingly marginal above this.  On several samples, the internal temperature of the A/D converter and other components was fairly high, but not alarmingly so, although this seemed to vary among samples (e.g. some seemed worse than others.)  Since thermal resistance can be characterized by a temperature rise, it makes sense that as the ambient temperature increases, so will the components by the same amount meaning that if the unit is in a hot location - or placed such that it will become warm (convective air movement across the heat sinks is restrictive or in/near the hot air flow of other equipment) then thermal stresses of the components also increase.

Again, the reader should be cautioned that the reported inconsistency between units (e.g. the efficacy of the thermal pad) may mean that the above advice may not apply to all units as some may have, say, a misplaced thermal pad or extra "slop" in the spacing between the board and the case which reduces the compression of the pad causing extra thermal resistance.

"Board slop"doesn't help: 

Figure 4:
Measuring the "board slop" in the mounting rails.  As noted
in the text, the board's looseness was nearly 1 mm - the far
extent of which exceeding the 5mm thickness of the pad.
Click on the image for a larger version.

On this latter point (e.g. "slop" in the board position) with the covers removed I measured a variance of 0.170-0.205" (4.32-5.207mm) from the board to the case due to looseness in the board fitting in the rail on one of my RX-888.  Of the three units that I have to measure, this was the worst - but not by much as the the photo (figure 4) from another unit shows.

Considering that the thermal pad is nominally 5.0mm thick, this means that the board MAY not be effectively conducting heat to the case if the gap is closer to 5.2mm.  Also considering the fact that the thermal pad will work better when it is compressed it would be a very good idea - if possible - to reduce this gap - more on this later.

I also observed that with the USB end plate fitted, it happened to push the board "down" (e.g. reduced the gap between the board and the case) by about 0.02" (0.5mm) and since this is the end of the board closest to the A/D converter chip, it likely reduces the gap by about 0.015" (0.38mm) owing to geometry (e.g. the fact that the A/D converter is located away from the edge.)  If desired, this fact could be exploited by adding a shim to the top of the USB connector and filing the bottom a bit to allow the end plate to push "down" on the board a bit, better-compressing the thermal pad and potentially reducing its thermal resistance. 

Figure 5:
The screwdriver tip points to where the end plate is pushing
down on the connector and board to reduce board-to-case
distance to better-compress the pad.
Click on the image for a larger version.
On the opposite end of the board, the RF connectors fit rather loosely in their mounting holes meaning that one could, in theory, move the connectors to the "bottom" of their holes and tighten the nuts on the SMA connectors.  This would not be advisable without adding a washer of appropriate thickness between the plate and the SMA connector as the connectors themselves are not right at the edge of the circuit board and firmly tightening the nuts would likely bend/break them loose.

Before getting out the file, however, I suggest considering the methods/modifications mentioned below to improve the thermal performance of the RX-888 (Mk2) in several other ways.

Ways to improve the thermal performance:

There are two ways to improve the thermal performance and reduce the temperature of the onboard components.

Add another heat sink and a fan

A "brute force" approach to this would be to move more air through and around the unit. using a small fan.  If you do this I would recommend two minor modifications:

  • Glue the heat sink to the A/D converter.  As noted earlier, the heat sink the A/D converter is held on by tape, but I would recommend that this be removed from the heat sink and the chip itself (using a bit of paint thinner or alcohol to remove residue) and it be reattached using thermally conductive epoxy rather than conventional "clear" epoxy.  This epoxy is readily available at the usual places (Amazon, etc.) but it should be noted that the gray (not clear!) "JB Weld" epoxy (available at auto-parts and "big box" stores) also has reasonable thermal conductivity and works quite well in this application.   Do NOT use an adhesive like "super glue" as it is not void-filling by its nature and it is unlikely to endure the heat.
  • Add a heat sink to the FX3 chip.  This chip - next to the A/D converter - should also be cooled and a small heat sink - such as that which comes with a Raspberry Pi heat sink kit - may be attached.  Again, I would recommend thermally-conductive epoxy rather than supplied double-sided sticky tape.

As for the fan mounting, several people have simply removed both side plates and fabricated the attachment for a small fan (say, 20x20mm to 30x30mm) on the side with the USB connector to blow air through the case on both sides of the board.  Others have temporarily removed the board from the case and put holes in "top" of the case (on the side with the labels) into which a fan is mounted.

Either of these will be quite effective - but since these are not passive cooling, the failure of a fan could result in excess heat if other methods are not also employed.

Improve passive cooling by using a much larger thermal pad

This is likely the favored approach as it does not depend on a fan which will have a defined useful lifetime, and the failure of which could result in immediate overheating in certain circumstances.  There are two parts to this approach:

Replace the thermal pad. 

At reasonable ambient temperatures I believe that the area of the external heat sinks on the RX-888 are of adequate size, provided that they are open for air flow and not placed in the heat exhaust of equipment and properly attached to the case - more on that shortly.

As noted, the thermal pad is seemingly marginal and it is only as large enough to draw heat away from the area immediately proximate to the A/D converter - an issue that may be exacerbated by the inconsistent board-to-case spacing mentioned above.  Improper placement of this pad will prevent it from conducting heat from the A/D converter - the major heat producer - to the case - and subsequent heating of adjacent components.

Figure 6:
A piece of 45mm x 65mm thermal pad on the bottom of the
board.  This piece is large enough to cover all heat-
generating components.
Click on the image for a larger version.
It is also likely that the thermal pad material supplied with the unit is of lower thermal conductivity than other materials that are available (to save cost!) so the use of better thermal material and a larger pad will draw more heat away from all of the heat-producing components on the board and conduct it to the heat sink.

A suitable pad material is the Laird A15340-01 which may be found at Digi-Key (link here ).  This material has roughly half  the thermal resistance (e.g. better thermal conductivity) of other common pad materials and it is suitably "squishy" in that it will form around components and help fill small voids as it does so.

Unfortunately, this material is somewhat expensive in that it's available only as a rather large piece - about $32 (at the time of posting - not including shipping) for one that is 22.8x22.8cm square - but this will modify several RX-888s - but even at the price of $32, it's still a reasonable price to pay for improved reliability of a $150-$200 device!  If you do this, it's recommended that you get with others to split the cost of the pad - but be sure to keep the pad - or any pieces that you cut from it - in a zip-bag or clean plastic cling film to prevent its surface from being contaminated with dirt and dust.  If you post this pad material to someone else, be sure to protect it between two pieces of cardboard to prevent it from being mangled.

Note:  Others have obtained 5mm thick thermal pad material from other sources (e.g. Amazon) and while it likely does not have as low thermal resistance as the Laird product mentioned, reports indicate that it works adequately - most likely a result of the larger size of this pad compared to the original, drawing heat away from the entire bottom surface of the board.

Figure 7:
The new pad, installed, as viewed from the
end with the USB connector, near the ADC
and FX3 USB interface chip.
Click on the image for a larger version.

A rectangular piece of thermal pad 45mm x 65mm will cover the bottom of the board where there are heat-generating components and ensure superior heat transfer to the case.  Since this material is a bit "sticky", it may be a bit difficult to get it installed as it will be resistant to sliding, but a very light coating of white heat-sink grease on the side of the pad facing the heat sink material will provide sufficient lubrication to allow it to slide as the board is inserted along its mounting rails.

Comment:  This process is fairly messy, so if you plan to add a connector for an external clock input, I would suggest that you do so at the time that you install the new pad as you will probably not to repeat the process unnecessarily.

Remount the heat sinks.

As noted earlier, the four heat sinks (to on the "bottom" side opposite the label and one on each side) are held on by double-sided paper tape.  It is recommended that these be removed - along with any tape residue (best done with paint thinner and/or alcohol) - and be reattached with thermal epoxy.

Figure 8:
An RX888 (Mk2) in the process of gluing on the side heat
sinks, using a vise for clamping.  Alternatively, weight may
be placed on the heat sink(s) while the epoxy cures to
compress it and squeeze out excess - but note that until it
cures that the heat sinks may slide slowly out of position
if one isn't careful.
Click on the image for a larger version.

As noted previously, the heat sinks do not fit flat with each other so  it would be a good idea to assure that the surfaces are reasonably to maximize thermal conductivity by drawing the case and the mating surfaces of the heat sinks across 800-grid sandpaper (using a flat piece of metal or glass as a substrate) - taking care to prevent metal particles from getting onto the board or inside the case:  It would be best to remove the board and do this prior to the installation of the new thermal pad and wash any such particles from the case before reassembly.

Once the mating surfaces have been flattened and cleaned, using thermal epoxy (or the gray "JB-Weld") reattach the heat sinks one-at-a-time - preferably by compressing them in a vice or with a clamp to squeeze out as much adhesive as possible.

It's worth noting that even if you don't go through the trouble of flattening the heat sink and the surface of the case, the use of a void-filling adhesive will certainly offer far more efficient thermal transfer than  the original double-sided paper sticky tape along with it s rather large air gap between the two surfaces.

Out of curiosity I measured the difference in temperature between the heat sinks stuck on with double-sided tape and the exposed portion of the case right next to the heat sink and it was found to be about 3-5F (1.7-2.8C) - surprisingly good, actually.

Before and after thermal measurements

Figure 9:
Two RX888 Mk2's with reattached heat sinks, ready for a 
bit of clean-up and final assembly.
Click on the image for a larger version.
Using a thermal infrared camera and verifying with a thermocouple, temperature measurements were made of various components with an RX-888 operating at 130 Msps at an ambient temperature of 74F (23C) after 10 minutes of operation.  The readings were as follows:

With the original thermal pad, end plates removed - heat sink cooling by convection only:

ADC:  175F (79C)

FX3 (USB interface): 155F (68C)

Capacitor near 3.3 volt regulator:  145F (63C)

3.3V Regulator:  170F (77C)

1.8V Regulator:  178 (81C)


With Laird 45mm X 65mm pad - heat sink cooling by convection only:

ADC: 145F (63C)

FX3: 130F (54C)

Capacitor near 3.3 volt regulator:  125F (52C)

3.3V Regulator:  145F (63C)

1.8V Regulator:  150F (66C)

Note:  There is another capacitor near the 1.8 volt regulator, but it is temperature cannot be readily measured while the board was installed in the case, but other measurements made outside the case indicates that its temperature was at least as high as that of the capacitor near the 3.3 volt regulator.

Results and comments:

The replacement of the original thermal pad with one that is 45mm X 65mm in size to cover the bottom of the board where there are active components has resulted in a very significant heat reduction:  As with all electronics, reducing the temperature of the components will increase the operational lifetime.

Considering that one can use - as a guideline - the temperature rise above ambient, we can make some estimations as to what will happen if the modified RX-888 (Mk2) is operated at a higher temperature.  

For example, if we consider 212F (100C) to be the maximum allowed case temperature of any of the components, we can see that with the original thermal pad, this limit would occur with the ADC converter at an ambient temperature of around 111F (44C) - a temperature that one could reasonably expect during the summer in a room without air conditioning.  In contrast, with the larger pad the ADC's temperature would likely be closer to 185F (85) in the same environment.

With a small amount of air moving across the heat sinks, their temperature rise would also be lower, further reducing internal temperature - and even though it isn't strictly necessary, it wouldn't hurt to use a small fan - even on a modified RX-888 (Mk2) to cool it even more, and feel confident that it will still survive should that fan fail.

Finally, I would again remind the reader that I consider the RX-888 (Mk2) to be an excellent-performing and extraordinarily flexible device and well worth extra trouble to make it better!

* * *

This page stolen from



Monday, July 31, 2023

A solid state replacement for an old radio's "vibrator" (Wards Airline 62-345)

Figure 1:
The front of the Wards Airline 62-345 with its rather
distinctive "telephone dial" tuning dial.
It's powered up and running from 12 volts!
Click on the image for a larger version.
Quite some time ago - a bit more than a decade - a friend of mine came to me with an old "Farm" radio - a Wards Airline 62-345.  This radio - from the 1930s - was designed to run from a 6 volt positive ground battery system  such as that which one might find in tractors and cars of that vintage.

How high voltage was made from low voltage DC in the 30's

As the technology of the time dictated, this radio has what's called a "vibrator" inside - essentially a glorified buzzer - that is used as a voltage chopper along with a transformer to convert the 6 volts from the battery to the 130-150 volts needed for the plates of the tubes within.  Not only did this vibrator do the chopping for the high voltage, but it also performed the duty of synchronously rectifying the AC waveform from the transformer as the pulses from it would naturally be in sync with the motion of the moving reed, briefly connecting the output of the transformer to the input of the high voltage DC supply when the voltage waveform from it was at the correct polarity.

These devices, as you would expect, don't have a particularly long lifetime as they are constantly buzzing, making and breaking electrical contact and causing a small bit of arcing - something that will inevitably wear them out.  Even if the contacts were in good shape, the many decades of time that have passed will surely cause these contacts to become oxidized - particularly since these devices are in rubber-sealed cans (to minimize noise and vibration) and the out-gassing of these materials is likely of no help in their preservation.

Figure 2:
The chassis of the radio.  The vibrator is in its original
can in the far right corner.
Click on the image for a larger version.
Such was the case with this radio.  Often, the judicious application of percussive repair (e.g. whacking with a screwdriver) can get them going and if the contacts are just oxidized, they will often clean themselves and work again - at least for a while.  In this case, no amount of whacking seemed to result in reliable operation, so a modern, solid-state approach was needed.

The solid-state replacement

As mentioned earlier, the job of the vibrator was to produce a chopped DC waveform, apply it to a transformer for "upping" the voltage and then use a separate set of contacts to perform synchronous rectification - and our solid-state replacement would need to do just that.  That last part - rectification - was easy:  Just two, modern diodes would do the job - but chopping the DC would require a bit more circuitry.

The owner of this radio also had a few other things in mind:  He changed it from 6 volts, positive ground to 12 volts, negative ground so that it could be readily operated from this more-common power scheme.  The change to 12 volt filaments required a bit of work, but since all of the tubes were indirectly heated, the filament supply could be rearranged - but some tubes had to be changed to accommodate different filament voltages and currents as follows:

  • Oscillator and detector:  This was originally a 6D8 (6.3v @ 150mA) and it was replaced with a 6A8 (6.3V @ 300mA).  Other than filament current, these tubes are more or less the same.
  • IF Amplifier:   The original 6S7 (6.3v @ 150mA) was retained.
  • 2nd Detector/AVC/1st Audio:  The original 6T7 (6.3V @ 150mA) was retained.
  • AF Output:  The original 1F5 (2.0v @ 150mA) was replaced with a 6K6 (6.3v @ 400mA).  The latter is a pentode, requiring a bit of rewiring and rebiasing to replace the original triode.
  • Magic Eye tube:   The original 6N5 (6.3v @ 150mA) was replaced with a 6E5 (6.3v @ 300ma) - which is also a bit more sensitive than the 6N5, giving a bit more deflection.

The 6T7 (150mA), 6A8 (300mA) and the #47 dial lamp (6.3v @ 150mA) are wired in parallel on the low side with one end of the filament grounded while the 6K6 (400mA), 6S7 (150mA) and 6E5 (300mA) are wired in parallel on the high side with one end of the filament connected to +12 volts.  You might notice a current imbalance here (600mA on the low side with 850mA on the high side) but this is taken care of with the addition of 30 ohms of resistance between the midpoint of the filament string and ground to sink about 200mA getting us "close enough".

He also did some additional rebiasing and other minor modifications - particularly for the rewiring of the AF Output from the original 1F5 to a 6K6 as he swapped a triode for a pentode - which was then  wired as a triode.  The total current consumption of the radio at 13 volts is 1.6 amps - a bit more than half of that being the filament and pilot lamp circuits meaning that about 10 watts of power is being used/converted by the vibrator supply and consumed by the idle current of the audio output and other tubes.

The other issue with the 6 to 12 volt conversion is that of the primary of the high voltage transformer:  This transformer is center-tapped with that connection going to the "hot" side of the battery (which was originally at -6 volts) - but what this really means is that there's about 12 volts from end-to-end on the transformer at any instant.  We can deal with this difference simply by driving the transformer differently:  Rather than having the center tap "hot" with the DC voltage and alternatively grounding one end or the other as the vibrator did we can simply disconnect the transformer's center tap altogether and alternately apply 12 volts to either end, reversing the connection electronically to preserve the original voltage ratio between primary and secondary.

This feat is done using an "H" bridge - an array of four transistors that will do just what we need when driven properly:  Apply 12 volts to one side and ground the other - or flip that around, reversing the polarity.

Consider the schematic below:

Figure 3:
Solid state equivalent of a vibrator supply.  This version uses an "H" bridge, suitable for
the conversion of a 6 volt radio to 12 volt operation as detailed in the text.
Click on the diagram for a larger version.

This diagram shows a fairly simple circuit.  For the oscillator we are using the venerable CD4011 quad CMOS NAND gate with the first two sections wired to produce a square wave with a frequency somewhere in the 90-150 Hz region - the precise value not being at all critical.  The other two sections (U1c and U1d) take the square wave and produce two versions, inverted from each other.

Figure 4:
The top (component side) of the circuit.  This is built on a
piece of phenolic prototype board.
Click on the image for a larger version.
The section of interest is the "H" bridge consisting of transistors Q1 through Q4 wired as two sets of complimentary-pair Darlington transistors.   Here's how it works:

  • Let us say that the output of U1c is high.  This causes the output of U1d to be low as it's wired as a logic inverter.
  • The output of U1c being high will cause the top transistor (Q1 - a PNP Darlington) to be turned OFF, but at the same time the bottom transistor of this pair, Q2, will be turned ON, causing the connection marked "PIN 1" to be grounded.
  • At the output of U1d - being low - we see that the bottom of this pair of transistors, Q4, is turned OFF, but the top transistor Q3 is turned ON causing V+ (12 volts) to appear at the connection marked "PIN 5".
  • In this way, the low-voltage primary of the transformer has 12 volts across it.
  • A moment later - because of the oscillator - the output of U1c goes low:  This turns off Q2 and turns on Q1 - and since this also causes the output of U1d to go high this, in turn, turns off Q4 and turns on Q3.  All of this causes "PIN 5" to now be grounded and "PIN 1" to be connected to V+ - thus applying the full 12 volts to the transformer in reverse polarity.

Also shown are D1 and D2, the solid-state replacements for the synchronous rectifier of the original vibrator.  While this could be a pair of high-voltage diodes (>=400 volts) we simply used half of a full-wave bridge rectifier from a junked AC-powered switching supply.  Finally, resistor R3 and capacitor C2 form a filter to keep switching noise and high-voltage spikes out of the power supply of U1 to prevent its destruction - a sensible precaution!

Now some of you might be concerned about "shoot through" - the phenomenon when both the "upper" transistors (Q1, Q3) might be on - if only for an instant - at the same time as the "lower" transistors (Q2, Q4) as the switching is done.  While this may happen to a small extent, it has negligible effect:  This circuit is efficient enough that no heat sinking is required on transistors Q1-Q4 and they get only barely warm at all.  Were I to build it again I might consider ways to minimize shoot-through, but this would come at the expense of simplicity which, itself, is a virtue - and since this circuit works just fine, would probably be not worth the effort.

Figure 5:
The bottom (wired side) of the circuit with flying leads
connecting to the original base socket.
Click on the image for a larger version.

These days one might consider building this same type of circuit using MOSFETs instead of Darlington transistors (e.g. P-channel for Q1 and Q3, N-channel for Q2 and Q4) and this should work fine - but the Darlington transistors were on hand at the time that this circuit was built and very easily driven by U1 - and the bipolar transistors are - at least in this case - arguably more rugged than the MOSFETs would be - particularly since there was no need to include a "snubber" network to suppress switching transients that might occur.  It's also worth noting that while standard MOSFET transistors would work fine for a 12 volt supply, you'd have to be sure to select "low gate threshold" devices to work efficiently at 6 volts or lower - something that would not really be an issue with the bipolar Darling transistors shown here.

This circuit is simple enough that it was wired onto a piece of phenolic prototyping board, snapped down to a size that will nicely fit into the original can that housed the vibrator.  To complete the construction, the top of the can - which was originally removed by careful filing and prying - was glued into its base using "shoe goo" - a rubber adhesive - keeping the board protected, but also allowing it to be easily disassembled in the future should modification/repair be necessary.

To be sure, the Internet is lousy with this same sort of circuit, but this version has worked very well.

What about the center tap version of the solid state vibrator?

You might ask yourself "what if we don't want to rewire a 6 volt radio to 12 volts?"  As noted previously, the boost transformer in the radio had its center tap connected to the "hot" side - which, in this case, would have been the negative terminal (because many vehicles had 6 volt, positive grounds at the time).  This circuit could be easily modified for that as you'd need only "half" an "H" bridge and the resistors driving the transistors would be changed to a lower value - perhaps 2.2k.  Depending on whether the it was positive-ground or negative ground, or whether the center-tap was grounded or "hot" - this would dictate whether you needed the PNP or NPN halves of the H-bridge.

(If you have a specific need, feel free to contact me by leaving a comment.)

* * * 

This page stolen from




Wednesday, July 19, 2023

Modifying an "O2-Cool" battery fan to (also) run from 12 volts

A blog posting about a fan?  Really?

Why not!

Figure 1:
The modified fan on my cluttered workbench, running
from 13 volts.
The external DC input plug is visible on the lower left.
Click on the image for a larger version.

This blog post is less about a fan, but is more of example of the use of a low-cost buck-type voltage converter to efficiently power a device intended for a lower voltage than might be available - in this case, a device (the fan) that expects 3 volts.  In many cases, "12" volts (which may be anything from 10 to 15 volts) will be available from an existing power source (battery, vehicle, power supply) and it would be nice to be able to run everything from that one power bus.


Several years ago I picked up a 5" battery-operated DC fan branded "O2 Cool" that has come in handy occasionally when I needed a bit of airflow on a hot day.  While self-contained, using two "D" cells - it can't run from a common external power source such as 12 volts.

Getting 3 volts

Since this fan uses 3 volts, an obvious means of powering it from 12 volts would be to simply add a dropping resistor - but I wasn't really a fan of this idea (pun intended!) as it would be very wasteful in power and since doing this would effectively defeat the speed switch - which, itself is just a 2.2 ohm resistor placed in series with the battery when set to "low".

The problem is that the fan itself pulls 300-400 mA on high speed.  If I were to drop the voltage resistively from 12 volts (e.g. a 9 volt drop) - and if we assume a 300mA current - we would need to add (9/0.3 = ) 30 ohms of series resistance to attain the same speed on "high" as with the battery.  The "low speed" switch inserts a 2.2 ohm resistor, and while this works with its original 3 volt supply, adding this amount to 30 ohms would result in a barely noticeable difference in speed, effectively turning it into a single-speed fan.  By directly supplying the fan with something close to the original voltage, we preserve the efficacy of the high/low speed switch.

Fortunately, there's an answer:  An inexpensive buck converter board.  The board that I picked - based on the MP1584 chip - is plentiful on both EvilBay and Amazon, typically for less than US$2 each.  These operate at a switching frequency of about 1 MHz and aren't terribly prone to cause radio interference, having also been used to power 5 volt radios and even single-board computers (such as the Raspberry Pi) from 12 volts without issues.

These buck converters can handle as much as 24 volts on the input and provide up to 3 amps output - more than enough for our purpose - and can also be adjusted to output about any voltage that is at least 4 volts lower than the input voltage - including the nominal 3 volts that we need for the fan.

An additional advantage is the efficiency of this voltage conversion.  These devices are typically 80% efficient or better meaning that our 300 mA at 3 volts (about 0.9 watts of power) would translate to less than 100mA at 12 volts (a bit more than a watt).  Contrast this to the hypothetical resistive dropper discussed earlier where we would be burning up nearly 3 watts in the 30 ohm resistor by itself!


One of my goals was to retain the ability of this fan to run at 3 volts as it would still be convenient to have this thing run stand-alone from internal power.  Perhaps overkill, but to do this I implemented a simple circuit using a small relay to switch to the buck converter when external power was present and internal power when it was not, rather than parallel the buck converter across the battery.

If I never intended to use the internal "D" cells ever again I would have dispensed with the relay entirely and not needed to make the slight modifications to the switch board mentioned below.  In this case I would have had plenty of room in the case and freedom to place the components wherever I wished.  In lieu of the ballast of the battery to hold the fan down and stable, I would have placed some weight in the case (some bolts, nuts, random hardware) to prevent it from tipping over.

The diagram of this circuitry is shown below:

Figure 2:
Diagram of the finished/modified fan.
On the left, J1 is the center-positive coaxial power connector with diode D1 and self-resetting
resetting thermal fuse F1 to protect against reverse polarity.  The relay selects the source of power.
Click on the image for a larger version.

The original parts of are the High/Low switch, the battery and the fan itself on the right side of the schematic with the added circuits being the jack (J1), the self-resetting fuse (F1), D1, R1, the buck converter and the relay (RLY).

How it works:

When no external power is applied, the relay (RLY) is de-energized and via the "NC" (Normally-Closed) contacts, the battery is connected to the High/Low switch and everything operates as it originally did.

External power is applied via "J1" which is a coaxial power jack, wiring the center pin as positive:  The connector that I used happens to have a 2.5mm diameter center pin and expects an outer shell diameter of 5.5mm.  There's nothing special about this jack except that I happen to have it on-hand.

When power is applied, the relay is energized and the high/low switch is disconnected from the battery but is now connected, via the "NO" (Normally Open) contacts, to the OUT+ terminal of the buck converter.  

Ideally, a small 12 volt relay would be used, but the smallest relay that I found in my junk box was a 5 volt unit, requiring that the coil voltage be dropped.  Measuring the relay coil's resistance as 160 ohms, I knew that it required about 30 mA (5/160 = 0.03) and if we were to use 12 volts, we'd need to drop (12 - 5 =) 7 volts.  The resistance needed to drop 7 volts is therefore (7/0.03 = ) 233 ohms - but since I was more likely to operate it from closer to 13 volts much of the time I chose the next higher standard value of resistance, 270 ohms to put in series for R1.

Figure 3:
Modification of the switch board.  The button is
the positive battery terminal and traces are cut to
isolate it to allow relay switching.
Click on the image for a larger version.
The diode D1 is a standard 1 amp diode - I used a 1N4003 as it was the first thing that I found in my parts bin, but about any diode rated for 1 amp or greater could be used, instead.  Placing it in reverse-bias across the input of the buck converter means that if the voltage was reversed accidentally, it would conduct, causing the self-resetting thermal fuse F1 to "blow" and protect the converter.  I chose a thermal fuse that has several times the expected operating current so I selected a device that would handle 500-800 mA before it would open.

Modification to the switch board

The High/Low switch board also houses the positive battery contact, but since it is required that we disconnect the battery when running from external power, a slight modification is required, so a few traces were cut and a jumper wire added to isolate the tab that connects to the positive end of the battery as seen in Figure 3.

Figure 4:
The top of the board battery board. The
connection to the Batt+ is made by soldering to
the tab.
Click on the image for a larger version.
Near the top of the photo in Figure 3 we see that the trace connecting end of the 2.2 ohm resistor has been separated from the battery "+" connector (the round portion) and also along the bottom edge where it connects to the switch.  Our added jumper wire then connects the resistor to the far end of the switch where the trace used to go and we see the yellow wire go off to the "common" contact of the relay.

In Figure 4 we can see the top of the board with the 2.2 ohm resistor - but we also see the wire (white and green) that connects to one of the tabs for the Battery + button on the bottom of the board:  The wire was connected on this side of the circuit board to keep it out of the way round battery tab and the "battery +" connection.

The mechanical parts

For a modification like this, there's no need to make a circuit board - or even use prototyping boards.  Because we are cramming extra components in an existing box, we have to be a bit clever as to where we put things in that we have only limited choices.

Figure 5:
Getting ready to install the connector after
a session of drilling and filing.
Click on the image for a larger version.
In the case of the coaxial power connector, there was only one real choice for its location:  On the side opposite the power switch, near the front, because if it were placed anywhere else it would interfere with the battery or with the fan itself as the case was opened.

Figure 5 shows the location of this connector.  Inside the box. this is located between two bosses and there is just enough room to mount it.  To do this, small holes were drilled into the case at the corners of the connector and a sharp pair of flush-cut diagonal nippers were used to open a hole.  From here it was a matter of filing and checking until the dimensions of the hole afforded a snug fit of the connector.

Figure 6:
A close-up of the buck converter board with the
attached wires and BATT- spring terminal.
The tiny voltage adjustment potentiometer is
visible near the upper-left corner of the board.
Click on the image for a larger version.
Wires were soldered to the connector before it was pressed into the hole and to hold it in place I used "Shoe Goo" - a rubber adhesive - as I have had good luck with this in terms of adhesion:  I could have used cyanoacrylate ("Super" glue) or epoxy, but I have found that the adhesive bonds of these tend to be a bit more brittle with rapid changes of temperature, mechanical shock or - most applicable here - flexing - something that the Shoe Goo is meant to do.

Because this jack is next to the battery minus (-) connector, a short wire was connected directly to it, and another wire was run to the location - in the adjacent portion of the case - where the buck converter board would be placed.

Figure 6 shows the buck converter board itself in front of the cavity in which it will be placed, next to the negative battery "spring" connector.  Diode D1 is soldered on the back side of this board and along the right edge, the yellow self-resetting fuse is visible.  Like everything else the relay was wired with flying leads as well, with resistor R1 being placed at the relay for convenience.

Figure 7:
The relay, wired up with the flying leads.
Click on the image for a larger version.

Figure 7 shows the wiring of the relay.  Again, this was chosen for its size - but any SPDT relay that will fit in the gap and not interfere mechanically with the battery should do the job.

The red wire - connected to the resistor - comes from the positive connector on the jack and the "IN+" of the buck converter board - the orange wire is the common connection of the High/Low switch, the white/violet comes from the "OUT+" of the buck converter and goes to the N.O. (Normally Open) contact on the relay, the white/green goes to the N.C. (Normally Closed) relay contact and the black is the negative lead attached to the coil.

Everything in its place

Figure 8 shows the internals of the fan with the added circuitry.  Shoe Goo was again employed to hold the buck converter board and the relay in place while the wires were carefully tucked into rails that look as though they were intended for this!

Now it was time to test it out:  I connected a bench power supply to the coaxial connector and set the voltage of my external test power supply at 10 volts - enough to reliably pull in the relay - and set the fan to low speed.  At this point I adjusted the (tiny!) potentiometer on the buck converter board for an output of 3.2 volts - about that which could be expected from a very fresh pair of "D" cells.

Figure 8:
Everything wired and in its final locations.  On the far left is
the switch board.  To the left of the hinge is the relay with the
buck converter on the right side of the hinge.  The jack and
negative battery terminal is on the far right of the case.
Click on the image for a larger version.
The result was a constant fan speed as I varied the bench supply from 9 to 18 volts indicating that the buck converter was doing its job.

The only thing left to do was to make a power cord to keep with the fan.  As is my wont, I tend to use Anderson Power Pole connectors for my 12 volt connections and I did so here.

As I also tend to do, I always attach two sets of Anderson connectors to the end of my DC power cords - the idea being that I would not "hog" DC power connections and leave somewhere to plug something else in.  While the power cord for the fan was just 22 gauge wire, I used heavier wire (#14 AWG) between the two Anderson connectors so that I could still run high-current devices.

* * *

Does it work?

Of course it does - it's a fan!

The relay switches over at about 8.5 volts making the useful voltage range via the external connector between 9 and 16 volts - perfect for use with an ostensibly "12 volt" system where the actual voltage can vary between 10 and 14 volts, depending on the battery chemistry and type.

Figure 9:
The fan, folded up with power cord.
The two connectors and short section of heavy
conductor can be just seen.
Click on the image for a larger version.
Without the weight of the two "D" batteries, the balance of the fan is slightly precarious and prone to tip forward slightly, but this could be fixed by leaving batteries in the unit - but this is not desirable for long-term storage as leakage is the likely result.  Alternatively, one may place some ballast in the battery compartment (large bolt wrapped in insulation, a rag, paper towel, etc.) or simply by placing something (perhaps a rock) on the top.  Alternatively, since the fan is typically placed on a desktop, it is often tilted slightly upwards and that offsets the center of gravity in our favor and this - plus the thrust from the airflow - prevents tipping.

This page stolen from