This is a fairly simple modification which can be installed/removed at will. It came about after giving a talk on my K2 to a local club. As part of the presentation I included a couple of plots of the SSB filter response. I thought that these were quite good, however, a member of the audience said, that in his experience better performance could be achieved from the filters. I then started looking more closely at what my KSB2 filter performance really was, and if it could be improved.
I began by plotting the frequency response curves by using a noise generator and WaveSpectra software. Having tried various audio analysis software, I find that I am using WaveSpectra more and more. It is not as easy to use as Spectrogram, but is free, and more importantly supports 32bit analysis. This gives a higher resolution and a lower noise floor.
To find WaveSpectra goto WaveSpectra. As this is a Japanese site, probably most of the screen will contain unrecognisable characters. Near the bottom of the page you will find WS131.LZH. Click on this and download the file into a suitable directory. The .LZH file is a compressed archive, so use a program such as WinZip to decompress it. The file you need is WS.EXE. Extract this file into your working directory and run from there. You will find it has a lot more options than Spectrogram, but then again it can do much more.
On the Elecraft reflector there has been much discussion on the KSB2 filter performance, much of it centering on blow-by. This is where signals bypass the crystal filter, causing a degradation to the receiver passband characteristics. I spent some time investigating whether this was an issue with my sideband filter. Although in any rig this happens (I notice it on my FT1000MP) the K2 filter did not seem to suffer unduly from this problem. I tried different screening including one between CC and CM, both on the top of the KSB2 board and on the underside, but observed no real improvement in the low level filter skirts.
I wanted to try and improve the steepness of the SSB filter sides. This is because I operate mainly on SSB contests where strong adjacent signals can activate the AGC and depress the overall receiver gain, making on frequency weak signals more difficult to copy. Using an audio filter such as the KDSP2 does not solve this problem as it is outside the AGC loop. I didn't know whether I could bring about an improvement, but it was worth a try.
My initial thoughts were to add more crystals to the 2nd IF Xtal filter, however after an exchange of mails with Wayne in which he pointed out the difficulties of making the new filter track with the different modes, I adopted plan "B". It was Bob G3PJT who suggested that replacing the noise blanker with a 4 or 5 pole filter. This has the advantage of being easy installation and removal. I have not found the present noise blanker to be particularly effective, besides which who wants to have a noise blanker in circuit during a contest !?
Having decided to make another crystal filter, I debated whether to make one from scratch using a new set of matched crystals from Elecraft, or try another route. If I started from scratch I would also need to design and manufacture a PCB. It seemed that I was re-inventing the wheel ! The solution when it came to me was remarkably simple, I would use an existing board - the KSB2. A post on the reflector produced a loaded board that was surplus to requirements. This had all the major items I needed.
I left the crystals in place, but removed most of the components that surrounded them i.e. diodes, res. packs, etc. The only other components left on the board were T1, T2, C5 and C11.
While the crystal were easy to access I arranged the grounding leads to be at the bottom of the crystal cases. My existing KSB2 filter was fitted with capacitors that gave a nominal bandwidth of 2.5kHz, so I fitted identical values to the new filter. To minimise the capacitive coupling between CC and CM I made sure that they were both oriented in the same direction. This should ensure that a signal plate in one capacitor is opposite a grounded plate (unless their manufacture is haphazard !). On T2 primary winding pin 4 needs to be grounded. The noise blanker is held in place by a screw into a metal pillar on the RF board. To ensure that nothing shorted out, and to provide a mounting point for an 8 pin 0.1" header plug I used a small piece of perforated board. Again, the photographs show the detail. The header is glued to the perforated board and miniature coax is used to make the filter input and output connections. The filter input connection is from the header pin 1 to the free end of C5, and the output connection is from the header pin 6 to the free end of C11. The coax shields are grounded to header connector pins 2 & 5. Make sure that header pin 3 (+12V) and pin 7 (Auxbus) are not connected to anything.
Although the performance of the Pre-Filter Version 1 exceeded my expectations, following comment from Elecraft and others on the reflector Version 2 was constructed. This version corrects a potential problem where the IMD performance is degraded due to incorrect termination of the Post-Mixer Amplifier Q22. On each side of the crystal filter an "L" shaped assembly made from single sided PCB material soldered at the edges was glued to the main board with Epoxy Resin. On the input of the crystal filter a -5db pad was fitted. Transformer T2 was removed and a 5dB JFET amplifier powered from the +8R rail was connected to capacitor CP. All components were soldered "Manhatten" style i.e in the air. Things were a tight fit, and required a steady hand. I had not really noticed any problems with Version 1, but Version 2 should now have the same IMD performance as a K2 without the noise blanker. The noise blanker input circuit does not appear to provide the best matching for the Post-Mixer Amplifier.
Plots of the response before and after show quite an improvement in the steepness of the filter sides down to 60dB with only a slight increase in passband ripple. There is little change in signal level on swapping between the noise blanker and Pre-Filter. I haven't had the opportunity to try the Pre-Filter in anger in any contests yet, however tuning around the bands weak signals now seem to be easier to copy when close to strong ones.