Full-Range IF Selectivity; The Hammarlund Variable Crystal Filter

[G6Tanuki]

The single-crystal-and-phasing-control at 455KHz was sort-of OK for CW reception but as noted in similar recent threads it's not that appropriate for SSB or DSB reception due to its poor shape-factor.

The balanced 2-crystal approach [replace the phasing capacitor with a second crystal spaced a couple of kHz from the first] works much better for AM/SSB. Tweaking the loaded-Q of the tuned-circuit the dual-crystal filter so it's mis-terminated works into gives you the opportunity to manipulate the practical bandwidth as also noted.

Q-multipliers have their place, but the disadvantage of using one to sharpen-up the bandwidth of a SSB/AM receiver for receiving CW is that a single high-Q circuit is prone to 'ringing'. Actually that's an issue with multiple high-Q circuits too...

Filters are a funny bunch, most people think of them as linear devices - under strong-signal conditions they can be anything but! A detail that seemed to escape some of the early designers of broadband front-ends where the first xtal filter could cause significant intermodulation if there was a bunch of strong signals being presented a long way outside the passband.

Also, there's much to be said for 'distributed' filtering; putting all your selectivity in one excellent filter at the front-end of the receiver, then following it with 100dB of wideband gain is silly.... those gain-stages will add wideband noise which intermodulates with the signal at the detector. Some supposedly smart radio-manufacturers didn't understand this [Icom, I'm looking at you!] and some of their radios were known for a high level of background noise. Putting a less-tight-than-the-front-end filter immediately before the detector to cut out the noise-sidebands is the answer here.

[Radio Wrangler]

If you plot the third-order intermodulation curve of a crystal filter, using closely spaced tones, You will see the expected 3:1 slope in the lower level regions, but higher up you will see rather erratic behaviour, not just deviation from the expected slope, but evidence of multiple modes and even hysteresis.

Old FDM systems shoved a few thousand phone channels as SSB signals stacked to fit the bandwidth of a coax cable or microwave link. Instruments to unravel these multiplexes and measure levels as well as other tests have to meet stringent distortion and intermod specs. For some time it seemed that crystal filters could not give the required dynamic range, and a generation of instruments were designed with entirely L-C filtering. These weren't single or double superhets, they were hex-superhets! IFs went lower and lower in stages. At each IF, the filters had to be good enough to block the image of the next mix-down, so it was compromises all the way down, the selectivity of one IF limiting how low the next could be (IE how close the image got).

Eventually, we did go to receivers with crystal filters once the Loveland Division had discovered that the a local Colorado firm could make some very special crystals. At South Queensferry we were looking into NPR tests - very aggressive and demanding. After trying many crystal filter firms (there were many back then) we found only the C.E.P.E arm of Thomson-CSF could do the deed.

The half-lattice circuit uses two crystals... look at that AR88 circuit, and the second crystal simply replaces the phasing trimmer. To a first approximation the C-zero case capacitances of the two crystals cancel and the response can be symmetrical. The crystals are cut to plant the series resonance of one crystal on the parallel resonance of the other. Controlling the spacing between series and parallel resonances of the crystals controls the bandwidth, so we are now closing the door on the idea of variable bandwidth through variable loading. The bandwidth is being cast in quartz!

So 2-crystal half-lattice is still only a 2-pole filter, so roll-off stabilises at 40dB/decade. If you want >80dB stopband, that still means monstrously wide skirts. So you add another and another maybe another half-lattice pair. You set the pairs for different bandwidths, around the same centre. In this way you can construct 4, 6, 8 or more order filters with the poles positioned to fit Butterworth, Chebyshev and even transitional filter designs.

It's a bit complex, and not only are all your crystals different, the differences have to be tightly controlled (Hark, hark, the cash registers are ringing up BIG bills)

So, selectivity is difficult, selectivity is expensive and is one of the main limitations on receiver design. DSP is now fairly cheap, but getting into digits with high performance brings its own difficulties.

Not having selectivity right before a mixer or demod allows noise on image frequencies to get mixed into the mixer output, so the noise effecto of preceding staes is doubled.

An Ex-RAF guy lecturing on receiver structures used the phrase "The wider you open the window, the more s*** flies in" So always look down your proposed receiver structure with a very jaundiced eye loking for unnecessary bandwidths on paths and asking awkward questions like what happens to the amplitudes of big off-chanel interferers?

It's not easy, but it can be fun.

David

[ms660]

A link to the QST article (QST Dec. 1938, mag. page 33):

https://worldradiohistory.com/Archiv...ST-1938-12.pdf

Lawrence.

[Bazz4CQJ]

Originally Posted by ms660

I found the Hammarlund that I had was better than the AR88D.

Lawrence.

Thanks for that comment. I am not trying to get in to a "which is best debate", but just to get feedback on the Hammarlund (hereafter "Ham") circuit, which seems to have been successful, but overlooked to some degree. Over the years, there have been quite a few magazine articles about upgrading the HRO, esp in SWM, I but I've never seen one where the IF filter was addressed, other than Peter's article in BVWS introducing the mechanical filter.

Thinking about what to try as an upgrade to the HRO, the tapped inductor (designated as both L34 and as TR4) in the AR88 looks like it would be harder to re-engineer than the switched resistors used by HAM, so I'm still inclined to try the HAM circuit first. It should be possible to build it within the metal box which housed the original filter, so not quite a "plug-in" replacement, but one that should not be too difficult.

B

[ms660]

Originally Posted by Bazz4CQJ

Thanks for that comment. I am not trying to get in to a "which is best debate", but just to get feedback on the Hammarlund (hereafter "Ham") circuit, which seems to have been successful, but overlooked to some degree.

The Xtal filter circuit in mine worked very well so far as I remember, the only fault that sticks in my memory was an OC resistor in the Xtal marker circuit.

Lawrence.

[Bazz4CQJ]

I wanted to simplify the circuit diagram of the filter so I've re-drawn it to British convention and also, I've left out the circuit to switch out the crystal.

Given that the circuit allows a quite wide bandwidth, it seems to me that for a comms Rx, there is no need to switch it out. Leaving out that feature simplifies the construction and will also reduce the capacitance around the crystal. Does that sound sensible?

B

[Radio Wrangler]

Originally Posted by Bazz4CQJ

Given that the circuit allows a quite wide bandwidth, it seems to me that for a comms Rx, there is no need to switch it out. Leaving out that feature simplifies the construction and will also reduce the capacitance around the crystal. Does that sound sensible?

The selectivity curves are given in the QST article which Lawrence has linked to.

Skipping the widest bandwidth (crystal switched out) The next widest one is 30dB down at the +/-3kHz points. The others are even narrower. So the widest you'd then have is going to be too narrow for enjoyable broadcast reception, and rather peaky in the audio band for clarity on SSB.

Looking at the curves, the Hammarlund has a reasonable bandwidth at its widest for AM speech, but not good for music. The rest are more tuned to Morse reception.

It all depends on what you want it to do. Of course, the IFT operating impedances and Qs may not be quite what the Hammarlund used, and the crystal specs may be different. You'll only know when you've tried it.

David

[Bazz4CQJ]

Yes, fair comment. The proof of the pudding will be in the eating! But I do note the comment in QST that they went to great effort to keep the capacitance around the crystal low, even making a special holder made out of the material "isolantite", a ceramic and now a tradename. I'll be trying to re-use the HRO xtal, and the one I have is the one with solder tags on opposite sides of the unit as distinct from the plug it one that many sets used. Given the tiny relays now available, maybe one of those could be used to short the crystal?

The HRO lives in a bedroom which also has a quite nice Roberts radio, so that could offer some music, but if I crave a a few tracks from Joni, it will be downstairs and go to my trusty NAD's .

B

[dtvmcdonald]

I have an SP-600 as well as a National NC-400.

They both have variable (stepped) bandwidth crystal filters. However,
the principles of the variability are just the reverse of each other in terms
of loading.

Both of mine were absolute nightmares to get working as spec.

I tested and adjusted the filters by feeding noise into the radio and hooking
the output of the filter stage into my SDR radio by connecting the SDR
input to one of those "clamp-on" black heat dissipation shields
stuck on the tube but not grounded.

The Hammarlund was easier. It still had its original crystal in it, which
apparently had changed only a little bit over the decades. Still, the
characteristics were not anywhere near in spec. In this case minor
adjustments to the three resistors R45, R46, and R47 and L37 while
rocking the phasing control got it right. However, in the end a 2 pF NPO ceramic cap had to be added across one side of the phasing control
to get a symmetric response with the phasing control set at its center.

The National set uses, as a said, a sort of "reverse" design as far as
impedance goes. That set had a replacement crystal and was amazingly wrongly working. It the sharpest position it was 6 dB down at +- 12 Hz.
Thats not a typo: it was twenty-four Hertz wide. It switches both resistors and capacitors. The same technique was used to adjust it, but each and
every part had to be changed, by humongous ratios to the original. In the end it met specs.

Thus, I believe that the claim that it needed a "special" crystal is correct.
That is, it had not only have a reasonably correct frequency, but had to have exactly the correct characteristics including correct equivalent inductance, series capacitance and parallel capicatnece, and Q.

[Bazz4CQJ]

Originally Posted by dtvmcdonald

I have an SP-600 as well as a National NC-400.

Thus, I believe that the claim that it needed a "special" crystal is correct.
That is, it had not only have a reasonably correct frequency, but had to have exactly the correct characteristics including correct equivalent inductance, series capacitance and parallel capicatnece, and Q.

Thanks for that; interesting and perhaps a bit worrying.

Key question, when you say to get them working to the spec, to what extent would you have been willing to accept something that worked "quite well", even if not in spec?. In terms of improving the old HRO 5, especially for use on telephony, almost any modification could be an improvement.

One specific question; can you say anything about the differential capacitor used in the Hammarlund filter? Mistakenly, I initially believed it was quite a large value, but the early manuals don't specify a value, and as it is effectively in parallel with one of the caps on the mixer transformer, I began to realise it could not be very big. A while later, looking at the manual for the SP-600, I saw a value of 3.5-8pF specified in there.

B

[Radio Wrangler]

If you stick with a single crystal filter, you can only have a single-pole response until you get far enough out that the IFT-created selectivity starts to be felt.

This single pole response will be just that, you can only change it's Q and that stretches the whole response in the frequency dimension. Your 3dB points get wider but the skirts get proportionately wider too, so you loose the ability to reject adjacent channels. You are tied to a skirt slope of only 20dB per decade change in offset from centre frequency.

So the shape factor of the resulting filter is rather poor. The Hammarlund and AR88 circuits achieve variable bandwidth by varying the Q and stretching the whole frequency response.

If you want better selectivity in terms of better audio bandwidth of wanted stations and no worse rejection of unwanted stations, then the only route is to introduce multiple poles to create the new bandwidth. This can be done with crystals at 455kHz, but it will take a lattice structure to do it, and all the crystals will need tight control of their motional parameters and case capacitance.

So if you take a single crystal filter and widen it to the point where it is similar in bandwidth to an AM-appropriate IFT strip, the skirts of the crystal filter will be much wider than the IFTs. I'm afraid it's just one of those things built into the maths behind filters.

Other 455kHz filter technologies like Collins' mechanical filters and Murata etc's 'ceramic' filters are multipole assemblies and can give far better wanted signal/unwanted signal performance than any single crystal filter can possibly give for AM and SSB bandwidths. The single crystal filter is really only suited to CW reception, and it works well enough to be useful at that.

I think you won't make a significant improvement to your HRO by modifying the single crystal half-lattice. I believe others have grafted in Murata/Toko ceramic filters to get better SSB reception.

These filters may in themselves have impedances best suited to transistor circuits, but impedances can be transformed. A ceramic filter could be embedded between a pair of lowish-Q tapped IF transformers to make it compatible with existing valve circuitry. Transforming Z down, and then back up again.

David

[Bazz4CQJ]

Easy bit first; take a look at

https://worldradiohistory.com/hd2/ID...OCR-IDX-50.pdf

where it says of the Ham.filter, “….with the crystal in circuit, in the broad position, reasonably good quality can be obtained on music transmissions”

Re ceramics, our earlier thread discussed it at length. The only report I know of somebody using them with valves is Peter’s article in BVWS, but he said it worked, but then went on to use the mechanical filter.

Jon thought about ceramics more, but has not proceeded with them.

B

[Jon_G4MDC]

It's still work in progress. The gear is in the shed at the end of the garden. It's winter and the garden is wet...and it's Christmas.

Instead I'm doing thought experiments to work out the best way of measuring the response of the filter / IFT matching before putting it in an actual receiver.

The present favourite is to let an oscilloscope imitate the 1st IF Amp Valve (the 30pF of scope input capacitance ought to be subtractable from the IFT tuning capacitance) while feeding it from a sig gen with ~2k series resistor as the source. Once the bandwidth and ripple look OK, meaning it's terminated right, then proceed with the input matching from the mixer anode. Not sure how far down the skirts it will be possible to see using the scope. I would hope that -20dB should be possible. The only working scope is a Telequipment D61a.

[Bazz4CQJ]

Originally Posted by Jon_G4MDC

It's still work in progress. The gear is in the shed at the end of the garden. It's winter and the garden is wet...and it's Christmas.

Instead I'm doing thought experiments to work out the best way of measuring the response of the filter / IFT matching before putting it in an actual receiver.

.

OK; that's good! Last time you commented on that, you thought you needed to use one or more FET's to do it; is that still the case?

Yes, I too am wondering how to test and set up the Ham. circuit. The suggestions I'm seeing on the internet are that, considering the importance of the impedance-matching, the best way is to install it in the set. Doing that would create problems of its own, so I'm wondering about building a test rig with perhaps a pentode at each end? Not sure.

However, I am now cutting metal (and non-metal) and I have a "drawings package" showing how I'm going build the Ham. circuit in the HRO filter box. That box turns out to be copper-plated steel (who knew?). I had a look at the HRO crystal using a sig-gen and a scope and confirmed that it is functional. Mine is the late one with the directly soldered connections as distinct from the early large cylindrical one with the 2 pins. I think the one with the soldered connections is "special"; much lower stray capacitance around the quartz. As far as I can find out, in the mid 1930's, there were only two cuts of quartz (X and Y) which differed only in their tempco. I think that the idea that Ham. had a special crystal almost certainly refers to the "crystal unit" rather than the quartz itself.

B

[Jon_G4MDC]

First I'm going to try to do it passively.

The 1st IFT has a tap fed by the crystal filter so trying that both ways round (it's not at the winding centre) will be a start. I suspect it will still be high of the wanted ~2k ohms. I will try resistive loading if it's needed but of course that will start to increase loss. How much that matters remains to be seen.

It's a guess but I think that maybe the whole winding is about 31 turns and the tap is at 4T / 27T. It's wave wound so you can't take the turns off to count them! I found the attached in my folder about it but I can't remember where it came from to give it proper credit! Maybe someone will recognise it.

I suppose really this should have been in the thread which was more about HRO than Hammarlund!

[turretslug]

There were certainly packaged filter assemblies from Toko (?) found in semiconductor tuners/radios that consisted of a modest ceramic filter connected between low-impedance coupling windings of input/output tuned windings. One of these could be adapted for its matching into a more sophisticated filter. Not sure what the maximum DC voltage rating of these is, though.

[G6Tanuki]

Yes I remember the likes of the Toko MFH41T that was sold by Ambit/Cirkit in the 70s; they were a popular upgrade to AM CB sets back then.

[ms660]

If it's any use to anyone, some general info I have on file for the HRO, specifically the HRO5T (Xtal hard wired in the phasing unit) for the transformer that couples the Mixer to the Xtal filter, and the coil that couples the filter to the 1st IF amplifier:

Coupling transformer from Mixer to Xtal filter:

Primary:

10/41 Litz....235 turns...3/16" wave wound...1.5mH...7.3 Ohms.

Secondary....Same as above.

Coupling coil to 1st IF amplifier:

10/41 Litz....297 turns tapped at 233 turns...3/16" wave wound...1.93mH total (297 turns)...1.48mH (233 turns)...10 Ohms and 7.2 Ohms respectively.

Lawrence.

[Jon_G4MDC]

Thanks for the gen about the mixer coil Lawrence. I reckoned that the two pies were identical. It's good to have confirmation of it.

My turns calculation on the grid coil seems a bit far out though. Should have realised that amount of inductance would need more turns than that. It was done from the dimensions of the pie and an estimate of wire diameter. Not made easier by being cotton covered.

[Bazz4CQJ]

Originally Posted by Jon_G4MDC

I suppose really this should have been in the thread which was more about HRO than Hammarlund!

If you recall, the last thread was on the subject of selectivity of vintage receivers in general and was actually a spin-of of a thread specifically asking about how to improve the selectivity of an Eddystone 840.

By chance the people who participated (as "customers") in that thread happened to be HRO owners. The guy with the Eddystone disappeared. That thread is now being continued, focussed on the ceramic filters, with you as the lead, but I'll certainly participate.

The Ham. design could be used in receivers other than the HRO, as indeed to could the ceramic filters.

B