Valve FM Receiver - Tuning Range

[murphyv310]

Hi.
The ubiquitous ECC85, the ECC88 was a great valve for VHF FM, which can be subbed in most cases occasionally the odd resistor needs a change of value.

[Jez1234]

Originally Posted by Radio Wrangler

It wasn't a case of British FM tuner designers being incompetent, their produce was as good as it needed to be and they kept prices down to levels where they dare publish them.

In the UK, the broadcast band was heavily managed, sparsely populated and planned to keep RF selectivity and overload requirements modest

America was much more of a free-for-all. Toners had to have great dynamic range for DXing stations in other cities while withstanding the power of local ones.

The UK had all the stations serving your area on the same site so signal levels were roughly equal, and any distant stations were mostly dupes of what you already got locally.

Europe was more like America, and so the Revox tuners for example do quite well.

That ws then but this is now and the UK has a heavily [populated band, radiated from transmitters in many more locations. We now need American style performance. 3 IFTs won't cut it for selectivity. Simple front ends lack dynamic range.

David

Indeed. I was not unaware of all that but was talking from the POV of actually using vintage tuners on today's crowded bands

[Jez1234]

Originally Posted by murphyv310

Hi.
The ubiquitous ECC85, the ECC88 was a great valve for VHF FM, which can be subbed in most cases occasionally the odd resistor needs a change of value.

ECC88 may sometimes need neutralising as a cascode, using a small air-cored coil between bottom anode and top cathode. See Troughline schematic changes when ECC88 was brought in as RF amp in the stereo version.

[Jez1234]

Originally Posted by Radio Wrangler

I think we are still in a window of opportunity as far as availability of things like dual gate MOSFETs and ceramic filters, but they are vanishing from catalogues. A really high performance design would probably use neither but they're needed to build something teaching the trade-offs of classic FM techniques.

David

Out of interest then, what would be your recipe for the ultimate FM tuner?

[E93AFAN]

Yes, I'd be interested too as a "newbie" to this area of radio technology.

[Radio Wrangler]

That's one of those bottomless, how far do you go, questions.

Varactor diodes are limited in Q compared to air-variable capacitors. This increases loss and reduces RF selectivity, so a very high performance front end would use a mechanical variable capacitor. To avoid LO tracking difficulties I'd move it with a stepping motor. With a microstepping controller, it won't need a gearbox. I have something like this working as an HF preselector. I'd have a self-alignment signal source, a crystal controlled comb generator so the microprocessor could tune a signal at several points in the range, and play with the RF tuning capacitor to peak the response, thus learning a group of settings to be interpolated to give a very well aligned front end. I might go so far as having low value varactors as fine tuners for each stage (not too much Q reduction) so RF stages could be peaked individually.

One extra section on the mechanical capacitor would act as pretune for the synthesiser VCO. Not having a varactor doing all the tuning of the VCO tank will improve its Q and that will improve phase noise. FM demod converts phase noise straight to noise noise.

RF gain.... Hmm. There are some quite decint MMIC amplifiers with respectable noise figures and capable of large signal handling. Internally they either have feedback or degeneration to set predictable gain and it helps with linearity. For the last bit, though, something discrete would be a little bit better, but we are chaseing that last bit. So I'd probably go for some CATV class bipolar transistors with transformer-based feedback around them. This gives the advantage of feedback-controlled gain and linearity without the thermal noise created in feedback network resistors. Look up noiseless feedback amplifiers and you'll see plenty by David Norton, Chris Trask and the Qubit corporation. You have to be crafty at 100MHz RF transformers on small toroids but these things work and probably yield the best dynamic ranges. They need to be good enough not just to handle the wanted signal range, they need to behave well to higher frequencies so you don't get spurious oscillation stability issues.

Mixers?

There are two possibilities, diode rings are an obvious choice. The high dynamic range versions need a lot of LO drive power. Dynamic range can be enhanced by three neat tricks: employing an LO driver that tries to look like a differential current source driving the mixer, secondly using a squarewave LO waveform because the rejection of large signals modulating the switching time of the diodes is a function of the dv/dt of the LO drive around the switching point of the diodes. Thirdly, running the LO at twice the frequency you thought of and having a fast ECL divide-by-two which will give a predictable 50:50 mark space ratio actoss the tuning range. Where's the point in making a well-balanced mixer then hitting it with a significantly asymmetric LO signal? Consider the mixer as an invert-notinvert switch and it needs to spend equal times in each state. There's more on these techniques in HP journal April 1982. This was pushing schottky diode ring mixers about as far as they could go. It was a lower frequency application but in an instrument designed to handle 18.6 MHz of thousands of FDM telephony signals all at once and measure channel noise floors and dynamic ranges at the same time.

Synthesiser...

The straight-forward PLL isn't too bad if you take actions to reduce jitter in the dividers. There are a few bolt on tricks worth playing. One of mine is to have multiple (many) duplicate fractional-N noise shaping dividers running in parallel, but offset in time and offset in the seeds for their pseudo-random fractional-N systems. It's in US patent 6509800. It allows wanted control influences around the loop to add more favourably than the addition of noise contributions. I've built ones of these with 8 dividers and phase detectors all trying to control the same VCO! sounds silly but it works. If you look at that patent on google, look at where it's been cited!

Another VCO phase noise trick is to cancel some of it. This is used in HP/Agilent/Keysight's best low noise synthesisers. A side channel mixes the VCO down to a lower IF where it is fed into a delay-line type pulse count discriminator. With a very clean crystal comb generator doing the mix down LO, this side channel discriminator demodulates the phase noise of the LO, and the result can be fed back to a varactor in the VCO tank to partially cancel the VCO phase noise.

IF selectivity

In ye olde days we used IF strips with IF transformers spread down a chain of amplifiers. This wasn't a bad idea. The intertwining of gain stages and selectivity stages is very good for dynamic range. If we put all of the selectivity ahead of all the IF gain, the selectivity protects the amplifiers from adjacent signals and possible overload/intermod which is good, BUT all that gain has no bandwidth limitation on the noise created in those amplifiers. FM discriminators are non-linear things and all that noise will convolve into the wanted signal's frequency space. B***er!

But if we put all that gain ahead of the selectivity, then the amplifiers see all the adjacent channel signals which haven't been rejected yet driving them into overload and intermod. It's good for noise, though!

Modern FM tuners went down the selectivity first route with it all in one block filter. This suits IF intermod quite well, but makes it difficult to keep IF noise contribution insignificant. Balancing these demands means a lot of late nights wrangling with modelling spreadsheets sufficiently well developed to model non-linearity and noise aspects.

The commercial ceramic filters are good at what they do, but aimed at mass market prices. LC filters give design freedom, lower loss and can be shaped better for linear phase characteristics (non-linearity in phase response is converted into distortion by FM demodulation)

So I'd go for LC filters and not do all of it in one block.

Discriminator

Easy choice. Pulse-count using a delay line to set the width of the pulses being integrated. Same sort of thing as mentioned earlier re LO phase noise cancellation. They have good linearity and good noise. They're also easy to set up.

One other area of concern: Thinl of limiting amplifiers in the IF as being driven into switching. You want the switching threshold to be AT the midpoint of the signal. Otherwise any amplitude variation on the signal will move the timing of the switching action. This gives us a conversion of any AM noise into phase modulated noise and phase mod noise is FM noise with a sloping frequency response. THis issue is called AM-to-PM and is the bane of wideband FM FDM links.

As we've now got to audio, it's the right place to stop. Both typing fingers now sore.

This is extreme 'what if' stuff, and I don't want to scare a newbie off! (I also know what an E93A is but that's not forum stuff)

David

[Jez1234]

Very interesting David! Use of Norton type amps at RF and of pulse count discriminator I correctly predicted but I was surprised to see diode ring mixers and LC IF cans there!

I would be surprised if by now there were not integrated Horrabin "H-mode" and Tayloe mixers with monolithic techniques used to equalise gate propagation delays etc hence OK to 100MHz+

[Radio Wrangler]

I missed a whole paragraph!

I'd intended suggesting the H-mode mixer as an alternative. The ones I did in the early nineties could easily equal the performance of 'ultimate diode mixers' with the only power taken being 25mW into a 74AC74 driving the gates. The arrangement was suggested to me by Ian Buckner, an engineer with a background in low frequency stuff saying 'what's wrong with analogue switches' So I had a think and the more I thought, the more I liked it. P R Rafuse at MIT had done a mixer with double-diffused DMOS FETs substituted into a ring where diodes would normally live and published a paper in 1967 or 1968. He'd missed a trick. Why use a ring, a circuit designed to as much as possible make up for the inconvenience of using a two terminal = 1-port device with a 2-port device? He'd missed the benefits of separating the controlled from the controlling.

Now I've done some serious messing with RF transformers making things like return loss bridges for balanced impedances etc. Doing a hexafilar-wound transmission line transformer for an H mode mixer was do-able. HP put a disclosure of the technique somewhere that would prevent us being locked out of using it while not at the time thinking it was worth patenting ourselves. Over a year later, Pat Hawker published the first things about this type of approach. We'd not known anyone else was active in the area. Colin and his employers found it couldn't be patented.

Around the early nineties I was musing on doing a 2-metre front end for repeaters. The antennae at that time were still on major broadcaster's masts and were being hit by tens of volts of other services. Tough for the receiver!. Cavity filters were de rigeur, but I wondered if a hyper tough front end could ease the filter requirements and reduce the filter loss, aiding overall noise figure. So yes, an H-mode style mixer was possible but then the amateur radio world lost the use of BBC/IBA masts and the need evaporated.

So, yes an H-mode mixer would be what I'd choose.

I'd also consider dual conversion. Dropping to a lower IF helps pulse count discriminators and as I've spoken on the advantage of not doing all IF selectivity in one block for dynamic range concerns, if the selectivity is in two blocks, there's nothing forcing them to be on the same frequency.

Ian suggested his 'analogue switch' mixer without realising just what it meant.. Everyone thought it was too obvious to patent, even after I had one on the bench producing silly TOI figures. It didn't look hard enough to be impressive.

If you're looking around, on that other FM tuner thread, I mentioned Bart McJunkin's cartwheel resonator VCO. Bart patented it as an oscillator. I reckoned it would make a handy high-Q and agile preselector filter. Another one the firm let slip.

David

[Jez1234]

Thanks David, most interesting as usual

Indeed some of the best ideas are the ones which make you think "nah... it can't be that simple... or somebody would have done that decades ago".

I'll see your hexafilar coil and raise you my octifilar (horizontally interleaved) primary for a SMPS transformer I made a few weeks back

I made a little widget to ease the process which could actually be of use to others.
I cut a "finger" of veroboard from a section at a top or bottom edge where there are 2 or 3 undrilled tracks, filed the copper from this section then drilled 8 0.6mm holes as close together as I could. The rests pretty obvious but something that was "obvious in hindsight" was that you can interconnect the windings as you need when "loading them" to the widget and then no matter what happens in winding the thing in ends with all 8 still in the correct order/phase/orientation at the finish.

[kalee20]

Thanks for tech tips David!

Distributing the selectivity amongst the gain blocks seems fundamentally the 'right' thing to do.

Discriminator... Pulse-counting should work, but I'm interested why you discounted a PLL, with a suitably linear VCO.

(Actually, I have a vague idea of combining AFC with demodulation, by having a fixed-frequency oscillator at, for sake of argument, 10.7MHz, feeding a phase detector, the other input being the 10.7MHz IF. Then the output of the phase detector controls the LO, 10.7MHz above the incoming RF.

So, the LO is forced to follow the deviation of the incoming signal, and the IF stays at 10.7MHz with just a very slight advance and retard in phase... the IFT's can all be narrow-band, and each stage high gain.

Obviously, the distortion will depend on linearity of the LO as a VCO, but in principle, it seems to me it'll work, and be better than a wideband IF chain letting the deviation go right through to the demodulator.

Originally Posted by Radio Wrangler

HP put a disclosure of the technique somewhere that would prevent us being locked out of using it while not at the time thinking it was worth patenting ourselves.

You mean, publishing in an obscure regional newspaper somewhere abroad, where you hope none of your competitors get to read it... but if someone else invents it independently and tries to patent it, you pull out this newspaper and invalidate their patent as an idea already in the public domain?

[Radio Wrangler]

Theoretically, it looks almost like you could have a very narrow IF, relying on the PLL to make the LO track and thereby cancel the FM on the incoming signal so a very narrow IF is all you need.

But there's a but. So the only remaining forwards path from the antenna to the audio stages goes through an infinitesimal width channel.... so how does any information get through that?

A few people have had a go at this one over the years, (some with household names) the superficial explanation of how it works is so attractive, but the reason why the IF has to have at least the audio baseband sort of width above and below the carrier is a bit harder to see, but a simple information theory analysis comes up with the equivalent of a perpetual motion machine.

Second problem is that you want good phase flatness across not only the audible range on the output, but also far beyond audible so that the vector addition in the stereo decoder of signal components from the sum and difference paths which were separated in frequency still give correct summations and nulling. So a PLL demod needs a lot more bandwidth than first appears and it becomes quite a chore.

THere was an equivalent proposal in the AM world about a century ago called "Stenode Reception" exploiting the new crystal filters which had been invented. The proposal was to make a very very narrow receiver IF allowing a great many more stations to be witted within the same band. The inevitable muffled audio being fixed by an audio stage with rising frequency response. Proposed by a chap called Robinson. Worth looking up.

Oh, yes, large corporations' legal and IP departments cultivate lists of obscure places. HP, Tek and R&S had signed agreements giving each access to the other's patents. It was less costly than trying to sue each others socks off. A consequence is that each needed to hold a reasonable wodge of current patents to make the deal look balanced. Any employee creating a new patent in a roughly relevant area got a thousand quid tax-paid and a little presentation certificate laser engraved in hardwood as it all helped to keep the deal balanced.

David

[kalee20]

Thanks for reply, David!

Originally Posted by Radio Wrangler

Theoretically, it looks almost like you could have a very narrow IF, relying on the PLL to make the LO track and thereby cancel the FM on the incoming signal so a very narrow IF is all you need.

But there's a but. So the only remaining forwards path from the antenna to the audio stages goes through an infinitesimal width channel.... so how does any information get through that?

Yes, I originally had in mind, an infinitesimally narrow channel... akin to the infinitesimally small voltage, current at a the virtual-earth point of an inverting amplifier with an op-amp. The info still gets through...

Originally Posted by Radio Wrangler

...the reason why the IF has to have at least the audio baseband sort of width above and below the carrier is a bit harder to see, but a simple information theory analysis comes up with the equivalent of a perpetual motion machine.

I was actually hoping to circumvent Carson's rule, that for FM, the bandwidth needs to be 2 x (fd + fm) with fd being deviation and fm being the modulation frequency - and reducing it to not much more than 2 x fm, which is the same as AM.

Originally Posted by Radio Wrangler

Second problem is that you want good phase flatness across not only the audible range on the output, but also far beyond audible ... So a PLL demod needs a lot more bandwidth than first appears and it becomes quite a chore.

I'd thought of that too, that overall IF response would need to be careful for the PLL to be stable - a bit like a feedback system where anything above a single-pole rolloff leads to stability issues, clever compensation, and ultimately compromised operation as a feedback system. So if I want the PLL to track at frequencies up to 15kHz, I'd have to ensure there is at most one 'funny' within a couple of octaves of that.

Looks like it's back to the drawing board...

[G6Tanuki]

Pragmatically, before setting out on a quest to identify the 'best' you need to define whast _your_ best is.

I've always been more interested in DX reception - hearing remote stations - rather than HiFi listening.

When you're interested in hearing stations in Belgium, France, The Netherlands, Germany - but your 8-element beam antenna points straight down the Thames Valley at the murderous 100KW BBC National stations broadcast from Wrotham, trying to 'listen through' their nuisance gets interesting.

Some ideas:

People deprecate varicap diodes in the front-end, saying they can never approach the Q of a traditional capacitor-tuned circuit. But pragmatically, if you are seeking to cover the entire 88-108MHz range, just how accurately can you set the tracking when using mechanical capacitors.

Look at what the civil-air-band guys are doing. They have varicap-tuning, with the tuning-voltage derived from something like a 10-bit digital-to-analog converter. That gives you 1024 discrete programmable tracking-points. And with a separate DAC for each of the 4 or 6 varicap-tuned circuits in the front-end, that's an accuracy that no clockwork capacitor can match.

Some people may be perplexed by the idea of 'aligning' a receiver front-end by entering numbers into a laptop and then saving them into the receiver's Flash-memory.

But it works.

RF amps: do you _really_ need one? If so, go for one of the noiseless-feedback designs. Use lots of Milliamps and a good supply-voltage. Some people will be freaked-out by the idea of a receiver front-end taking 50mA at 24V and needing heatsinks... but it gets good dynamic range.

Mixers: double-balanced diode package-mixers have been available for 50 years. MD108. Drive them hard, preferably with as near as possible a square-wave to get switching action..

And remember to terminate them properly.

Local-oscillator: the cheap availability of GPS-disciplined 10MHz reference sources mean you can easily build a LO that's probably as frequency-accurate as the transmitter. Just beware of odd frequency-products from some synthesizers.

"AFC" is for people who don't have access to a stable reference source. Pragmatically, when doing DX listening it's a no-no, it will happily drag your receiver off to a strong adjacent-chennel station when your intended station fades out.

IF stages; get good selectivity in the filter immediately following the mixer, but as noted upthread, a 'tail end' filter at the end of the IF strip is also really necessary to exclude the noise-sidebands introduced by the 100-or-so dB of gain in the IF strip.

Don't feel you have to be bound to convention and use 10.7MHz as the IF; the UHF mobile-radio world went to 21.4MHz and so there are some good filters available at that frequency.

Detectors: yes, extract the I and Q components and then the rest is software; you'll need this anyway to decode the RDS station-idents [whose introduction in the last decade of the last millennium was a real boon to help us identify the distant stations we were hearing via tropo or Sp.E propagation].

Also - as an idea - Band-II is 20MHz wide; do you _have_ to tune it in one span?

Pragmatically, you could split it into four 5MHz-wide slices and so optimise the L:C ratios of the signal-frequency tuned-circuits. If you're not into varicap-tuned digitally-programed stuff you could perhaps use a 4- or 5-chamber stagger-tuned helical filter for each 5MHz chunk?

[Radio Wrangler]

Originally Posted by kalee20

So if I want the PLL to track at frequencies up to 15kHz, I'd have to ensure there is at most one 'funny' within a couple of octaves of that.

15kHz tracking would still strip the difference channel off and leave mono.

If you want stereo, you need it to be faster tracking than you'd first guess in order to have the difference path in the right phase to get good separation.

David

[Jez1234]

My own priorities would be hi fi rather than DX but as we are talking ultimate then it should be excellent in all regards. Getting really good results in stereo, with no background noise, weird warbles or multipath distortion when reception conditions are less than optimum is the ultimate test which most FM tuners fail, and where things like the much vaunted Leak Troughline with external decoder become pretty useless IME (yes even with selective dual gate mosfet low noise RF amp in front of it).

I don't recognise the existence of witchcraft such as SDR etc

[Cruisin Marine]

Originally Posted by Radio Wrangler

You would not believe the difficulty and testing needed to get an aviation NAV receiver design to adequately reject 107.9 while trying to receive a localiser signal on 108.1. You've only got 200kHz and you need to work with signals in the -90dBm region with good enough freedom from interference to meet navigational accuracy requirements.

One high up the FM band transmitter has a mast close to the coast, with the main approach to a large airport going right past it. Turning onto and acquiring localiser takes them almost over it. So yes, the worst case scenario does exist.

Doing a straight forward FM broadcast receiver looks quite inviting after that.

David

Sounds nigh on impossible. I guess digital processing techniques were employed?

[Radio Wrangler]

Originally Posted by Cruisin Marine

Sounds nigh on impossible. I guess digital processing techniques were employed?

Not really. All the intermodulation/crossmodulation and other nasties happens in the front end, long before the signal gets digitised. DSP is helpless to do anything about it.

If you don't have any analogue selectivity, the DAC has to absorb all in band and out of band signals hitting it simultaneously while remaining close to perfectly linear. The gain/loss from antenna to DAC has to be scaled such that the peak voltage created as the sum of all signals is within the linear region, and then the noise floor still has to be good enough to get the required S/N ratio with the lowest specified signal. That kills the concept of just connecting an antenna to a DAC.

It's not quite impossible but other things have to be sacrificed to pull it off. Later stages down the IF along with the demods did get gobbled up in DSP.

Welcome to the world of Dark Arts, aka RF design. I used to look like Harry Potter, scar oon wrist, not forehead.

David