Saturday, July 19, 2014

Replacing the filters in the Drake SP75 Speech Processor

A few months ago I fired up my old Drake TR-7, getting on the air with some friends on 40 meters after not having used the radio for a while.

Everything seemed to work fine, at least until I switched the audio through the matching Drake SP75 Speech Processor as some of those on frequency were having a little bit of trouble hearing me on the noisy band and I started getting the following helpful comments:
Figure 1:
The front panel of the Drake SP75 speech processor.


"That sounds terrible!"

"Your 'lows' are completely missing!"

"Your audio sounds 'restricted'"

"Turn it off!"

I obliged, of course, but I also knew that in the past I could switch in the speech processor, setting it to a very low level of clipping, and no-one could really tell the difference between it being switching in and switched out, so I knew that something had definitely changed!

Later, I used some available test equipment (computers and software) to see what had changed, setting up the following:
  • Using another computer (a netbook) I ran the Audacity  (link) program, a free, open-source audio editor.  Using that program, I generated 5-10 minutes of white noise and set it to play back that white noise as a loop.
  • The generated white noise from that computer was fed into the "Tape" input of the speech processor with the audio level set just high enough to properly drive it.
  • The audio output from the SP75 was fed into another computer running the Spectran (link) program - also free - which does audio analysis.  Another program that would work, but is more difficult to use with a much steeper learning curve, is "Spectrum Lab" (link).
The result of this was that I was able to compare the "flat" white noise input to the speech processor by the netbook with the audio spectra coming out of the SP75.  Ideally, the SP75 would not appreciably "color" the audio - that is, the the frequency response of the SP75 should be pretty "flat", not rolling off either the lows or the highs, at least in the frequency range used for speech (e.g. 100-300 Hz to 2700-3000 Hz or so).  Unfortunately, I can't seem to find the screen capture of that spectral plot or else I'd include it here.

By this point, I was fairly sure that I already knew the answer - and the above technique of "sweeping" the audio passband using white noise verified it:  The "low end" audio frequencies (below approximately 700 Hz) were being rolled off significantly - by 6-10 dB and more, explaining why my audio sounded so bad!

How the SP75 works:
Before we go on, a few words on how the SP75 works.

This is a combination AF/RF speech processor and it works by first routing the input audio through an XR2216 audio compressor chip.  Then, the audio is double-sideband modulated at around 459 kHz, filtered to produce a lower sideband signal using a pair of 455 kHz ceramic filters, RF clipped, filtered by another ceramic filter, and then demodulated back to audio.

By applying the clipping at RF, the resulting audio is devoid of harmonic distortion since the harmonics would occur at around 900 kHz instead of within the audio range itself.  The only real casualty being that there is now intermodulation distortion, but since the human voice largely contains sounds consisting of noise bursts (consonants) or voiced sounds with harmonics (vowels) the introduction of intermodulation distortion on speech, alone, has relatively little effect on the perceived quality and intelligibility if the clipping process is not taken to extreme.

By applying both audio compression - to assure a consistent amount of RF produced for the SSB modulator - and RF clipping, the "best of both worlds" in terms of audio processing can be applied in terms of improving the "peak-to-average" ratio for speech while minimally increasing the amount of perceived distortion.

What had gone wrong:

Because of the loss of low frequency audio, I figured that one (or more) of three things had happened:
  • One or more electrolytic capacitors in the audio path had dried out and decreased in value causing the loss of low frequency response.
  • The BFO, nominally at 459 kHz, had gone off-frequency and caused the audio passband through the filters to shift.
  • One or more of the 455 kHz ceramic filters had gone bad.
With the white noise applied to the input and still using the Spectran program to check the audio spectra I went though several of the audio test points noted in the manual, applied them to the audio input of the computer running Spectran and observed that at least to the balanced modulator, the audio was completely flat, ruling out the likelihood that a capacitor had gone bad.

I then fired up an SDR (Software Defined Radio) - an RF Space SDR-14 - and started probing around inside the SP75, noting that the BFO was, in fact, where it should have been:  within a few 10's of Hz of 459 kHz, ruling out the second probability of the above.

Connecting the input of the SDR-14 to test point 11, through a 2k resistor to minimize circuit loading, a location after all of the filtering and clipping and on the output of an amplifier stage, I centered the SDR on the passband - while still sending white noise through the SP75 - and looked at the resulting display and saw that it was anything but flat, indicating that one or more of the ceramic filters in the unit had, in fact, gone bad!

Identifying the ceramic filters:

In looking at the filters themselves they were clearly made by Murata, marked with "CFW455" followed by what looked like an "I6" printed in white ink while the schematic diagram simply called out a part number of "CFW455I".  In doing a bit of research on the web, I determined that the Murata "CFW455I" had the following specifications:
  • Center frequency:  455 kHz
  • Input/Output Impedance:  2 kohms
  • -6dB bandwidth:  +/- 2 kHz
  • Stop bandwidth:  +/- 7 kHz (at -50dB)
  • This device was a 6 pole filter
What was also apparent was that this particular filter was no longer being manufactured, so I started looking around for a replacement.
Figure 2:
The new filter (left) and the old filter (right). 
 All but one of the leads lines up with the original.
(When taking the picture, I cleverly rotated one of the filters
by 180 degrees, so keep that in mind!)
Click on the image for a larger version.

What I did NOT want to do was use "new-old" stock because of age-related degeneration with these parts.  Typically, these parts have silkscreened, silver-plated electrodes on the surfaces of their ceramic elements, but even though these are usually fairly well sealed, they gradually degrade for whatever reason, either due to slow corrosion of the potting compound that protects them, ingress of moisture from the environment, due to electrolytic degradation due to chemical reaction and/or voltage applied to their terminals, or maybe just the degradation of the potting compound and the plastic in which they are encased.

Whatever the reason for their degradation, I decided that I did not want to get "new" 10-20 year old parts and risk having them be out of spec!

In perusing the various catalogs I noticed that Murata still made a part that was electrically identical - the CFWLB455KJFA-B0, available from Mouser Electronics - so I ordered some.

Installing the replacement parts:

Figure 3:
New holes that need to be drilled into the board to accommodate the different pinouts of the new filters.
Note that this picture was taken before the old holes were completely cleaned of solder.  On the far right
hole, carefully avoid the adjacent trace - both when drilling, and when later soldering the jumper.
Click on the image for a larger version.

I'd ordered the CFWLB455KJFA-BO filters knowing ahead of time that while they were electrically identical, they were NOT mechanically identical, so with the new filters now in hand I set about modifying the SP75 circuit board after carefully removing the three original Murata ceramic filters using both a "solder sucker" and plenty of "Solder Wick" (tm).

Figure 4:
The trace at the center filter position (FL2) that
inevitably lined up with one of our newly-drilled holes.
Click on the image for a larger version.
As can be seen in Figure 2, the new filter is slightly smaller than the old one - and the pinout is slightly different, as well, but fortunately there is only ONE pin (the "output" - but these filters are bilateral, so it doesn't matter which is used for the input or output) that is actually in a different physical location which means that we need to drill just one hole for each of the filter locations.

Referring to Figure 3, above, you will note that the new hole is in line with an intersection of two imaginary lines drawn from other pins, so the new hole just needs to go, as shown in the picture, "above" the now-abandoned hole and inline with the "input" pin.

Of course, Murphy has to intervene as shown in Figure 4 where the extra hole drilled for FL2 ended up going right through through a trace on the top side of the circuit board.

Fortunately, we have the technology (a soldering iron, solder, a knife and wire) to relocate this trace and get around this problem,  literally!

Both ends of the trace that ran under the original filter were sliced with a sharp knife and the original trace was heated with a hot soldering iron so that it lifted off the board.  The "bloody ends" of the interrupted trace were then scraped clean of the green coating and a short piece (some #30 wire-wrap) of wire was soldered into placed, used to route around where the filter would be placed as depicted in Figure 5.
Figure 5:
The removed and re-routed trace using a short
piece of #30 "wire wrap" wire.
Click on the image for a larger version. 
Having done this, the board was now ready to receive the three new filters.

Because only the lead with the drilled hole does not match the original pinout, they may be (mostly) soldered as normal.  For that "other" lead, a short piece of wire - a trimmed component lead, for example - may be used to make the connection to the original, now-unused hole to the new pin as seen in Figure 6, below.

Figure 6:
On the bottom side of the board, the installed filters and the jumpers to the leads that connect
to the positions with the newly-drilled holes.
Click on the image for a larger version.

Meanwhile, on the top side of the board, the filters look like this:

Figure 7:
The new filters as viewed on the top side of the board.
The rerouted trace (the yellow wire) may be seen just to the left of the middle filter.
It is interesting to note that these "CFWLB455KJFA-B0" filters are actually marked "W455I"!
Click on the image for a larger version.

Getting the SP75 back into working order:
Firing up the SP75 after replacing the filters I noticed immediately that its audio didn't sound right - very "tinny", even worse than before.  Putting the white noise back into its input and connecting the SDR-14 to TP-11 I noticed immediately that the 459 kHz BFO frequency was entirely outside the passband of the 455 kHz filters.

What had happened?

From what I can tell, one of two things might have changed:
  • These new filters (CFWLB455KJFA-B0) are slightly narrower than the original CFW455I ceramic filters used by Drake.  In this scenario, the BFO and the edge of the audio passband would have been "moved" entirely outside the filter.  I have no way to know this for certain as the only filters I have for comparison are the now-30-year-old original filters.
  • The more likely scenario:  The original Drake filters were marked "CFW455I6" - a designation that doesn't seem to be correlate with anything in a catalog that I could find.  Perhaps the "6" indicates a center frequency of "456" kHz?  If this is the case, that would imply that the original filters were specially-selected for the higher center frequency and, perhaps, had matched bandwidths.  Based on what I was seeing, having the passband of the filter shifted up 1 kHz to 456 kHz would have put it in about the right place.
In either case, the original 459 kHz BFO frequency would not be suitable for the new filters, so how to change the BFO frequency?  The BFO frequency was originally set with a quartz crystal - a rather expensive component to have custom made - but there are two easy alternatives:
  • Slightly reworking the oscillator to use L/C components such as a 455 kHz IF "can" (transformer) as frequency-determining elements.  This would, at the very least, involve adding a series DC-blocking capacitor were this route taken. A bit of care would need to be taken to assure that this arrangement was temperature stable to within a few hundred Hz over the expected frequency range.
  • Using an inexpensive 455 kHz ceramic resonator - also available from Mouser.
I chose the latter since I had several of those, already on hand, plus they had the double advantage of being quite "pull-able" over the range of several kHz and they are fairly frequency-stable, likely to move only a few hundred Hz, at most, over the temperature range that one might experience.  As for the original 459 kHz crystal:  I have wrapped it in paper and plastic tape and secured it inside the SP75 case in case I need it for some reason in the future.

Figure 8:
A Murata 455 kHz ceramic resonator used in lieu of the original 459 kHz crystal and a 180 pF capacitor in parallel with the
frequency trimming capacitor - both components being mounted on the bottom of the board.  As noted in the text, the value
of this fixed capacitor was determined experimentally using the methods described in the text.
Click on the image for a larger version.
Simply dropping the 455 kHz resonator in place of the 459 kHz quartz crystal yielded a tuning range of about 460-463 kHz (the frequency range will vary depending on the nature of the resonator) so I had to "pad" C44, the 7-62 pF tuning capacitor with a 180pF capacitor - the value having been experimentally determined - to get it into the correct frequency range for the filters.  What I ended up needing for my filters was a BFO frequency of 458.7 kHz - easily within the tuning range of the 455 kHz ceramic resonator, but your specific BFO frequency may vary, depending on the ceramic filters that you end up with.

There's no real reason why LSB (lower sideband) must be used when picking the BFO frequency as these filters are symmetrical:  USB would have worked if the BFO frequency had ended up in the right (or wrong) place.
If you use an L/C network for setting the frequency, pick the frequency that gives the best results using the methods described below.  It so-happens, however, that ceramic resonators are easier to move up in frequency than down as this requires just series capacitance, so using a "high side" BFO and LSB is just easier in this case!

How to determine the correct BFO frequency for your filters:

To determine the correct BFO frequency I used the same method that I'd used to determine that the original filters had gone bad in the first place, that is:
  • Insert a white noise source - at just high enough audio level to drive the SP75, but low enough to avoid any overload or clipping - into the input of the processor.  The correct level is that which is just high enough to make the front panel "audio" light turn on solidly, and then just a little more.
  • Using a program like Spectran to observe the audio spectra, note the "flatness" of the audio output taken from the speech processor and fed into a computer.
By varying the BFO frequency one can see the effects of the filter's bandpass:  Too low a BFO frequency and the upper edge of the filter starts to cut off the low audio frequencies (the processor generates lower sideband, remember!) and too high, the intermodulation products of the clipping can start to affect audio quality if both sidebands are recovered and demodulated.

Because these ceramic filters are considered to be "low cost" they do have a bit of intrinsic ripple (their specifications are for +/2 dB of ripple)  and they do not have a "brick wall" response, so don't expect a superior "shape factor" - that is, a very abrupt cut-off - but rather a fairly gradual cutoff over the span of several hundred Hz or a kHz.  If you are a purist, you can order several extra filters so that you may pick and choose which one(s) give the best, overall response - but note that the circuit board cannot take very much soldering/unsoldering, so you would want to install sockets or some other temporary connections were you swapping filters in and out frequently!

By carefully adjusting the BFO frequency, one should be able to get a fairly flat frequency response down to 200 Hz or so and up beyond 3000 Hz, fully encompassing the frequency range of any transmit audio source that you'd be likely to use!

Consider the result below:

Figure 9:
The "sweep" of the audio output, using the "Spectran" program, of the SP75 after replacing the filtersusing
a white noise source on the input and sampling the audio on the output.  This is about as good as
one can expect using inexpensive ceramic filters like this, but listening  on the output with an audio
amplifier, it sounds quite good.
This "sweep" was produced with the Spectran program analyzing a white noise input
and averaging over 96 samples to help "smooth" out the result.  Even so, the "roughness" is still evident due to
the fact that we are, in fact, measuring the spectral energy of white noise as our original signal source!  Not shown in this
"sweep", the actual frequency response of the SP75 extended to nearly 5 kHz with gradual rolloff above 3 kHz.
Click on the image for a larger version.
Figure 9 shows a pretty flat audio output - within a few dB - from the SP75.  As noted in the caption, above, listening to an audio source inputted through the processor via an external amplifier, it does sound pretty "flat" to the ear.  As can be seen, there is a bit of roll-off below 1200 Hz and this is entirely due to the ceramic filters themselves, but this could, in theory, be corrected with a simple R/C network:  The roll-off below 100-200 Hz seems to be intentional by the designers of the SP75 and occurs mostly in the output stage, although a bit of it is, in fact, from the "edge" of the ceramic filters' response.

Wrapping it up:

Overall, I'm pleased with the result, even though the project was a bit more involved than I'd expected it to be.  Up to a clipping level setting of 6-9 dB, there is hardly any noticeable distortion added to the audio - just as it used to be when the SP75 was new!

Of course, a speech processor is one of those things that should be used sparingly.  Under normal conditions with good signals it is probably not needed at all and when conditions start to get a bit rough, the added compression - if not taken to a ridiculous level - it should add more "punch" to a signal than it would degrade the audio due to excess compression, clipping, distortion and/or coloration.  This particular processor's "clipping level" control goes all of the way to 20 dB - a ridiculous amount that yields results that may be intelligible, but are likely to be unpleasant, so that setting should never be used in any but the worst possible band conditions - if even then.

It is worth paying very close attention to the SP75 manual in setting up the input and output levels for the SP75 for the microphone that you plan to use. Unfortunately, if you use several different microphones with the SP75 and they each have wildly different output levels, things get very complicated as only the microphone on which the SP75 was originally set up will be driven properly unless you have outboard attenuators on the microphones to assure that they all have about the same levels!
If the input level is too high, there may be too much audio compression in the XR2216 stage while if too low, the efficacy of the processor itself is reduced.
Also, the output level control should be set so that the audio level is the same when the processor is switched out (e.g. bypassed) and in with clipping set to 0 dB.

This page stolen from