Digital Audio Filter modules for Vintage Rx

[Bazz4CQJ]

A current thread concerns improving selectivity in vintage receivers.

Back in 2017, Phil (G4SPZ) drew my attention to audio DSP filters, see (for example) https://www.sotabeams.co.uk/dual-ban...dules-ssb-cw/. - (be sure to play the video) and this https://www.sotabeams.co.uk/variable...odules-ssb-cw/.

Has anyone experience of adding one of these to a vintage Rx?

B

[Jon_G4MDC]

It will be of limited use added in the audio chain.

It would be excellent to have that response in the IF instead. That way the unwanted signals will never get to be demodulated.
After demodulation the unwanted sideband is just upside down monkey chatter that goes straight through any audio filter along with the wanted one.

I see a very similar looking dsPIC development board using dsPIC33FJ128MC804. I think the CW/SSB filters are probably based on loading a particular set of code on it. One version of the PIC claims the DAC can do 500ksps or another 1Msps. That might digitise a 455kHz IF. Now you can do I&Q trickery and select sidebands digitally and the the slow 100ksps DAC just outputs the audio. I quite like the idea of this...

I once put the HF IF Input of a DSP TV Dongle behind an HRO. It worked but the digital side was too slow. Signals came out about 1/2 second late. It was an absolute nightmare to tune on the dial but fine if you tuned around using the IF. I wonder how much delay the dsPIC approach would have.

[Radio Wrangler]

Such filtering is too late in the gain profile of the sorts of receivers needing it. If you use the excellent audio filter to allow you to listen closer to large signals, then those large signals will be inside the IF bandwidth but outside the audio bandwidth and will be overloading the IF mangling the small target signal.

Under easy situations, audio filters can sound quite good, but dynamic range considerations wipe out their benefits just when you need them.

The other approach with DSP filtering is to have a low-gain set before the digital filter. The ADC now has to handle the full dynamic range the set must handle at the input. If you try relying on much AGC in the analogue stages, you run again into similar dynamic range troubles.

An alternative low gain stratagem is to do a direct conversion receiver, but this has no discrimination of sidebands for SSB or CW use and FM is right out. So to get round this a quadrature direct conversion receiver will give some sideband discrimination, but not much. If you used precision components and got I and Q channels matched to 1% you get 40dB sideband discrimination, 1 degree error in phase produces about the same limitation. It'll work, but 40dB is poor in terms of skirt depth or image suppression. So you can't approach the capabilities of a normal receiver with mode-specific analogue filters.

There are some impressive general coverage receivers with DSP filtering, but they use low gain analogue stages, rather good ADC devices, and they still use analogue superhet structures with appropriately narrow crystal filters in a high IF.

There are also sets which digitise the whole HF range and do digital down conversion and selectivity. Hilbert transformed filters handle the phase shifting for quadrature paths in the digital domain, and by keepin the whole I/Q split, processing and recombination race hazard in digits, plenty of precision can give plenty of sideband discrimination and image rejection. The ADCs in these sets are critical, and ideally some preselection is needed to make their life less stressful. These sets work and are very agile, but don't quite hold their own against the best hybrid analogue superhet/DSP IF hybrids.

(Got a talk to give on this stuff on Wednesday!)

David

[Jon_G4MDC]

If we said a vintage RX has a 455kHz IF with a nose bandwidth of say 10kHz in total then the ADC has already had most of the strong interferers stripped away by the IF selectivity of the original set. It has done the preselector function.

In most cases the signals in the unwanted sideband are only about the same as the wanted. To get them down 20dB would be heaven. I appreciate this is not the ideal approach but I still think the improvement would be well worth having.

The worst thing might be the unwanted signals can still modulate wanted ones by ALC action. That might be a reason to use manual gain control.

[G6Tanuki]

One thing digital audio processing _IS_ rather good at is reducing the ear-shattering 'sharsh' you get on an empty HF frequency; great for when you're monitoring a frequency waiting for a net to start.

You can use such a filter as a sort of digital squelch; algorithmically it's not too hard to distinguish between white/pink noise and the human voice. You can also do some cunning tricks; I remember even in the 80s some of the last-generation military FM radios used a digital demodulator incorporating a deliberate signal path delay, which totally removed the traditional burst of noise you would normally get between the other party releasing his PTT and your squelch realising the signal had gone away.

[Bazz4CQJ]

OK, so back so ceramic filters as possible ways forward.

I've just dug out my 98%-built Haigue wobbulator. Bought at a silent key sale, it's nicely put together but just lacking the 4 coils, which I believe are available from Spectrum. The plan then is to create some kind of 'filter assembly' in to which I can fit one of the many alternative ceramfils and look at what they might do in the HRO.

Also, when I complete construction of that QST Rx (the design is now chilled in my mind) I will make that so that alternative ceramfils can easily be put in/taken out of that for comparison.

B

[Jon_G4MDC]

More looking into the dsPIC device.

The specs say the max sample rate for 12 Bit mode is 500kHz which implies max signal bandwidth of 250kHz and this is the stated max for the ADC. It's not clear whether it could be operated in 2nd Nyquist zone to digitise a 455kHz signal. One would imagine that should be possible.

If not then the 10bit mode has max sample rate of 1.1Msps and corresponding ADC max signal bandwidth of 550kHz. This should digitise about 10kHz worth of 455kHz quite easily.

The IF of the popular RTL-SDR DTV dongles, which give a fair performance as receivers on VHF and UHF, has an 8 bit ADC. I reckon there might be something in this approach. 2 Bits more and with the IF selectivity of the original receiver in front.

[Bazz4CQJ]

Well, the price of the first filter I linked to is £30 (other products will probably be available).

So "if the proof of the pudding....", then I think you should try one out Jon .
Even if it is not the solution, probably an interesting product to have?

B

[Jon_G4MDC]

I have done some C programming to make VFOs from the Si5351 and Si570 linked to other Microchip PIC development boards.
I'm just not confident my skills are up to this one - there is a big knowledge gap.

On the other hand I thought the same about the VFO projects to begin with so it just might take a long time HI

It's true that it is not a massive investment and I much prefer to add a little board between the last IF and audio stages (which could be bypassed by a simple switch) than to hack up the crystal filter box.

Putting the RTL-SDR dongle in there (which I considered at one time) would have meant mounting a USB socket on an HRO. That would cross a line and as already reported there were big issues with delay.

[Bazz4CQJ]

Well, we have ceramic filters in the background and now DSP.

I'm (slowly) doing two things in pursuit of the ceramics;

i) re-vitalising my wobbulator
ii) will make provision in my QST Rx to act as a test bed for ceramics.

I'm going make it possible to easily swop the ceramic filter in the QST (perhaps a HC6/U socket or a B9A socket). There may be a limitation here in that it will be the overall IF chain that determines the BW, so testing on one rig may have a degree of uncertainty go with it.

Last night, I took out my (newish) oscillating or omni saw, and cut the end off of the 455kHz (25kHz BW) IF filter from my dear old Pye Vanguard. It's the thing that's painted pale blue and the size of a 20 fag packet. Not sure what I expected to find inside, but was disappointed to see about 8 (or 10?) ferrite pot cores (with tuning slugs) and a host of silver mica caps, on two PCB's. Not sure what use I might make of those? I think I got them out undamaged.

B

[Jon_G4MDC]

Later ones of those (Westminster era) were reported to contain ceramics instead.

Red were channel spacing of 50kHz, Dark Blue 25kHz and Pale Blue or sometimes plain aluminium 12.5kHz. Generally they had a nose BW about 60% of the channel spacing. I believe it was possible to swap capacitor values around to convert one to another but I have never heard of anyone who actually did that.

[Bazz4CQJ]

As you can guess, it's built fairly tightly and making changes would need care. As I think it's come out without any damage, I wondered if I could make use of it as a "reference point" on the wobbulator against which to compare other components (probably not) ? It had been sitting doing nothing for the last 40+ years. It was packed internally with foam rubber which had begun to turn to goo, but was still at a point where I got most of it off with a paint brush. There are actually 10 cores. I've stripped the Vanguard carcase of pretty much everything except the 03-20 valve section, it's driver and the inverter, thinking that they might make a modest amplifier for a handheld 2m FM rig. Other than that, the case itself (it has both lids) might make a new home for something.

B