Re: Sherwood SE-3 Sync Detector mod for ham use

[k4kyv]

Do those demodulate in true DSB mode? 

Some of the so-called "sync detectors" I have seen advertised and reviewed are little more than SSB product detectors that eliminate one sideband or the other by the phasing method.  In other words, it is essentially the old "sideband slicer" concept with a PLL carrier lock feature added.  You could do almost as well by simply copying an AM signal using any receiver equipped with a product detector, in SSB/CW mode.  The only advantage of these devices is that the BFO locks onto the carrier to eliminate the tuning error. 

A true DSB sync detector receives both sidebands at the same time, and locks the BFO exactly on frequency and in phase with the original carrier, so that you get the full advantage of coherent double-sideband reception plus the advantage of demodulating with a product detector.  In most sync detectors the lock is achieved using a PLL to lock onto the existing AM carrier.  But there is also the Costas Loop, which uses the relationship between the mirror-image sidebands to establish lock, ignoring the carrier altogether.  In fact, that is the only kind of detector that can properly demodulate DSB suppressed carrier signals.

I seriously doubt that the Sherwood SE-3 uses the Costas loop, but I am wondering if it really demodulates in a true DSB mode.

Are they still being sold new, and how much does (or did) one cost?

[w3jn]

Nope, the SE-3 doesn't use a Costas loop, nor are the sidebands selectable.
It is a true PLL sync detector.

BTW just because a sync detector uses the phasing method to select sidebands doesn't mean it isn't a sync detector.  The GE YRS-1 is one such example.

The signal is split in two and feeds two 6H6s, which are fed with BFO signals 90 degrees out of phase.  The DC output of the I 6h6 feeds a reactance tube, which phase locks the BFO to the incoming signal.

The I and Q channels each go thru phase shifters and a matrix network cancels the unwanted sideband.

My SE-3 (the later ones may differ) is similar to the sync detector project in the early 90's ARRL handbook.   Uses a NE604 and several NE602 mixers.

[k4kyv]

The I and Q channels each go thru phase shifters and a matrix network cancels the unwanted sideband.

So what is this "unwanted sideband" business? 

If there is always an unwanted sideband when it is demodulating DSB, it sounds to me like it is actually working  like a sideband slicer.  There shouldn't be any "unwanted" sideband, since the detector should be demodulating both sidebands simultaneously with each one contributing 50% of the audio voltage at the output, just as in the  case of a regular envelope detector.

According to the Costas article in the 1956 IRE Proceedings, a true DSB synchronous detector can be configured to reject the interference on one sideband or the other while both sidebands are still contributing to the detector output.  But if the only option is to simply blanket reject everything on the USB or LSB, without the option of receiving both sidebands of the desired signal, it is basically copying AM as a SSB signal with a PLL BFO, and is not a true synchronous detector.

[w3jn]

True, Don, I guess I was referring to the sideband reject feature which is very handy if you have QRM above or below.

A Costas loop won't get rid of in-passband interference, as far as I know.  The Costas loop as you noted detects the audio phase difference of the I and Q channels and the error signal drives the BFO.  The matrix is identical regardless of whether the BFO sync comes from the audio or the carrier.  It's also possible to receive USB and LSB in different channels, and feed them into a stereo amp for a quite interesting effect.

THe SE-3, and others like it, are true synchronous detectors in that they sync with the carrier.  A different approach, but still synchronous.

[WA1GFZ]

I wonder if you could make this automatic with a monitor on the phase locked signal if you can get at it. Say it loses lock for over .5 seconds then the loop is made wider until it sees a lock again. Then it switches to tight loop mode. Sounds like you could do it with bus switches or analog switches. I built a synthesizer once that worked like that.

[w3jn]

Don, after re-reading your post I think I see where your confusion comes from.

The audio output of a Costas loop, or a sync detector like the YRS-1, or even a sideband slicer like the Central Electronics, is a function of how the I and Q channels are combined in the matrix output.  The matrix can be configured to reject USB, reject LSB, or combine both for USB+LSB. 

The CE slicer has all such options, but its BFO isn't phase-locked - it's free running.  It would probably be fairly easy to convert the CE slicer to a sync detector -either Costas or phase locked to the carrier.

Frank - that's certainly one option, but I coulnd't hear much, if any, increased distortion with the wider loop.  It would be nice to automagically switch between PLL and envelope detection if lock is lost.

[k4kyv]

Since sometimes with selective fading you may lose carrier but retain both sidebands and at other times sideband information may fade while the carrier is retained, a syncing system that simultaneously sampled both the sideband phase info and the carrier, and used the information from one or both to set the carrier would more reliably maintain lock than either the Costas Loop or the carrier PLL system used alone.

[w3jn]

The time constant of the PLL is slow enough to ride thru loss of the carrier.  The slowest time constant of the SE-3 is about 10 seconds (the FLUTTER position).  This mode is for copying signals with significant phase distortion due to ionospheric disturbances, according to the manual.  I suspect a well-designed Costas loop would use a similar slow time constant to ride thru loss of one or the other sidebands.

In actual use I haven't had it lose lock on a fading signal, even listening to some of the PW tropical band broadcashers around 4900-5000.

You sound outstanding all the time, Don, but the other night when we had that QSO the SE-3 really drove home the fact that your audio is extremely clean and undistorted.

[AB2EZ]

Questions about synchronous detectors

For synchronous detection of AM, I think one would be more concerned (in terms of subjective performance) about frequency "lock" (i.e., within .1 Hz) than phase lock. Is that correct?

If so, and given that broadcast stations (as opposed to vintage amateur stations) have very tight frequency control, wouldn't you want the sync detector (when listening to a broadcast station) to stay right on the frequency it was last locked to (assuming the sync detector has a very stable frequency reference) in the event of a carrier drop out? I.e. you would push the "search and lock" button only when tuning in a new station. When the carrier is present, the detector would phase lock. When the carrier drops out, the detector would keep its locally generated, surrogate carrier frequency right where it currently is.... until the carrier returns.

When listening to a vintage amateur transmitter in the presence of selective fading, but no significant interfering carrier(s), I would think you would use a carrier discovery algorithm whose search range increases the longer the carrier has dropped out. If the dropout is less than 1 second, the algorithm might keep the frequency of the surrogate carrier right where it is. Beyond that, it might begin to look for the carrier over a wider search range (e.g., +/- 50 Hz).

When listening to an amateur transmitter with a very stable frequency... I would think you would revert back to the algorithm used when listening to broadcast stations.

When listening to a vintage amateur transmitter in the presence of a significant interfering carrier... I would think that you might fall back on yet another algorithm.

So... wouldn't a modern synchronous detector... which uses a demodulation method that comes to life on the combination of a digital signal processor and a general purpose microprocessor (software defined)... have much more sophisticated and much more effective algorithms than any vintage or even relatively recent sync detector?

Stu

[w3jn]

Stu - no, you need phase lock.  .1 Hz difference will cause an annoying buildup and fade of the audio due to periodic incoherence with the carrier.

Due to the slow PLL time constant, the sync detector will stay pretty much on frequency during no signal intervals.  So if someone comes back and is almost (but not quite) on freq the SE-3 will immediately lock.

You don't need to hit the lock switch during tuning intervals at all if you're not bothered by heterodynes as you tune around.  Just tune close to the signal and the SE-3 will lock right in.

If you're offset tuning a signal (to escape interference, or to put the whole BW of your filter towards one sideband) the SE-3 has a switch whereby it will maintain lock (ie speeds up the PLL time constant) as you tune.  As soon as a signal comes on frequency, the PLL will attempt to lock to it.  How fast it does so is entirely dependent upon the PLL time constant, and how far off frequency the sync detector's LO is from that signal.

Stu, you're overthinking this with all these algoritms    In practice, the stock SE-3 works very well if everyone in the QSO is within +/- 150 Hz.  If they're a KC or two off, then the PLL time constant needs to be shortened so it can capture the off-freq signal quickly.  The mod described in the Receivers section of the Online Handbook accomplishes this well.

In a typical QSO you don't even notice the modified SE-3 locking/unlocking.

[k4kyv]

Questions about synchronous detectors

For synchronous detection of AM, I think one would be more concerned (in terms of subjective performance) about frequency "lock" (i.e., within .1 Hz) than phase lock. Is that correct?

No, for proper demodulation of double-sideband, the carrier must be exactly in phase (plus or minus perhaps a few degrees), or 180º out of phase with the original carrier.  The phase of the audio output from each sideband at the receiver detector is a product of that sideband and the carrier.  The audio output from each sideband  must be properly in phase with that from the opposite sideband, in order for the two to be additive.  If the carrier is out of phase the amplitudes of the two sidebands will not totally add up, and the phase modulation present in each sideband will not  completely cancel out, leaving a combination of amplitude and phase modulation, with a distorted envelope. 

If the carrier is rotated exactly 90º, the amplitude modulation products of the two sidebands will completely cancel, leaving a signal that is purely phase modulated.  That's how Armstrong designed his first prototype FM broadcast transmitter; he generated a DSB signal at a very low frequency using a balanced modulator, then re-inserted a carrier in quadrature (90º out of phase) with the original carrier, then applied the appropriate audio pre-emphasis to the audio that modulated the carrier to convert the signal from phase modulation to frequency modulation, and finally multiplied the FM signal up to the operating frequency.

The phase relationships between carrier and sidebands, both for SSB and DSB can easily be demonstrated with rotating vectors.  This information used to appear in the old ARRL Single Sideband for the Radio Amateur, the Collins SSB Handbook  and numerous other radio engineering publications.  I don't know if it still appears in any of the ARRL publications or not.  Also, Steve WB3HUZ (IIRC), has demonstrated this with an animated vector diagram.  Maybe Steve would be willing to post it again.

Synchronous detection is not the same thing as chopping off one sideband of the AM or DSBSC signal and then demodulating it with a re-inserted carrier as a SSB signal.

Since there is only one sideband with SSB, the phase relationship between the original and reinserted carrier does not matter.  For perfect reception all that is needed is for the re-inserted carrier to be precisely on frequency with the original.  Unfortunately, this often does not happen with amateur slopbucket, and is one of the reasons so many SSB voices sound like Donald Duck.

Commercial SSB and DSB suppressed carrier systems may transmit a "pilot carrier",  reduced to about 20 dB below the p.e.p. of the signal, to give the receiver's PLL something to lock onto.  Technically, this is reduced, not suppressed carrier transmission.

The amateur radio slopbucket wisdom developed over the years has been that voice quality does not matter, so precise reinsertion of the SSB carrier has not been considered important and amateur operators have become acclimated to hyellowey sounding audio.  Only now, with the advent of "ESSB" and audio quality adjustments on some of the newer rigs, is this long-standing indoctrination beginning to be questioned by the amateur SSB community.

[Steve - WB3HUZ]

So... wouldn't a modern synchronous detector... which uses a demodulation method that comes to life on the combination of a digital signal processor and a general purpose microprocessor (software defined)... have much more sophisticated and much more effective algorithms than any vintage or even relatively recent sync detector?

Well, yes, it surely could. To JN's point, it may not be necessary. One interesting idea that could be done with DSP is one I heard from Bacon many years ago. The idea is to chop each sideband up into many small frequency bins. Then each bin on each sideband is compared and the "best" one is chosen and used for the actual audio output. The tricky part would be coming up with the "best" measurement and implementing it with DSP. But if this could be done, you could almost always get clean, interference free audio.

[WA1GFZ]

I2PHD Ver .99 is the best sounding but only about 80 Hz wide BW.
Flex has a much wider range but doesn't sound as good.
I2PHD winrad is still being modified. It went wider than 80 hz and doesn't sound as good. Alberto is taking comments form users.
Me, I think it could be made automatic with a delay to drop back to an envelope detector until it locks again with a wide range then ratchet tighter as time goes on.
All in the delay you are willing to live with. No universal way so never likely to please everyone. My TCI/BR 8174 does a good job in DSP AM but the widest bandwidth is only 6 KHz.

[AB2EZ]

Frank
Steve
Don
JN

Thanks all... excellent information.

Stu

[Bacon, WA3WDR]

I took I and Q from an analog sync detector and did a complex FFT on each, then cross-mixed the real and imaginary from each bin with the real and imaginary from the other bin.  That gave me complex upper sideband bins and complex lower sideband bins of 256 to 1024 (512 was usually fine) little frequency bands each, depending on the length of the original FFT.  In a frequency range of 0 to 20 KHz, 512 bands would be 20,000/512 or about 40 Hz wide each.  Then I compared the two sets of narrow little bins frequency by frequency, and decided which one was higher in amplitude, and I assumed that the one that was higher in amplitude was higher because there was a heterodyne or noise on that side, so I took the information from the narrow little bin on the other side, and I saved that, and I went to the next frequency, and so on.  Then I took the single complex bin set that came from that, and turned that into time-domain audio.  The DSP could do the FFT and also do that, in real time.  This worked pretty well, and gave me about 40 dB of garbage suppression.  I could have gotten more with better phase accuracy.

What I liked about this approach was that a heterodyne never even made it to the speaker for a microsecond, it was dumped before being turned into audio.  This would dynamically comb-filter crap from both sides simultaneously and choose the audio from the side with the least audio in each of the 512  40 Hz wide bands.  And it could be 1024 20 Hz wide bands too, although I didn't notice much difference.  Yes, I would hear every selective fade, but it would dump the heterodynes and the SSB noises, and general side splatter was rejected the same way.  It would pick and choose which part of which sideband to listen to, and it did it automatically in real time.  There wasn't very much delay either, maybe 50 milliseconds, as I recall a 1024 long FFT would have twice as much delay as a 512.

To deal with selective fading, I recommend circular polarization antennas, and dynamic receive signal polarization tracking if possible.  Unfortunately I got really sick and tired of diddling with it, and then I moved, and started doing other things.  I keep thinking of getting back to this, especially since SDR is starting to pop up in the stuff I do now.

   Bacon, WA3WDR

[AB2EZ]

Bacon

A very interesting concept. I don't know if you had ever considered patenting it... but I think it is one of the most innovative concepts I have seen... and I have chaired the patent committee in a large corporation, and served as an "expert witness" in a number of patent litigation cases. Probably too late for a US patent, and definitely too late for many other countries... but still a really neat approach!

*In the US, once you publicly disclose a concept, you have 1 year to file a patent application. In some other countries, you cannot file a patent application after public disclosure.

Separately... once you synthesize the single sideband composite... does it make any subjective difference if the "carrier" you use to demodulate the composite is 0.1 Hz off from the true carrier frequency?

Stu

[WA1GFZ]

Stu,
I always thought the XOR phase detector on a VCXO would be a good approach because if the signal drops out the XOR goes to 50% duty cycle putting the VCXO in the middle of the tuning range. Placing the signal near the middle of the tuning range would require just slight tuning to phase lock. The VCXO may only tune a few KHz anyway keeping the system tight.  Unlocked it would be no worse than a BFO at or close to zero beat. The VCXO could run at 4X frequency so I/Q LO could be generated with flip flops.
Then there is the Racal method using a MC1357 FM demodulator to generate a carrier without a PLL.

[AB2EZ]

Frank

I agree with the advantage of setting up the sync detector so that the surrogate carrier will stay at a frequency that is  close to the real carrier frequency if the real carrier drops out.

For a PLL using a relatively simple implementation... of the type that would have made sense for consumer and amateur radio applications prior to (roughly) 1980... tuning the receiver to a point where the local oscillator's control signal is close to zero, even when the loop is phase locked, sounds like a good idea. You need some kind of indicator to help you do this (or else you have to open the loop, keep the local oscillator on,  and tune for zero beat)

With modern, highly stable (when free running) local oscillators, and "digital" methods for generating and storing the VCXO control signal... I think that there is no longer need to tune the incoming signal to the free-running local oscillator frequency. If the phase comparitor stops putting out a correction signal (because the carrier has dropped out), the VCXO control signal will just maintain the value it had just prior to the carrier drop out.

Stu

[WA1GFZ]

Stu, all you need is an indicator on the XCXO control voltage No signal an XOR will be sitting at 50% or 1/2 way. then just tune the incomming signal to the same spot. This could be done with a meter or Bar display. The Cubic R3030 has a cool meter function that does this to center the carrier in the passband.
The only problem is when you are in a round table and a station is a bit off frequency. Still a station should be close enough for a VCXO to lock. 

[w3jn]

Still a station should be close enough for a VCXO to lock. 

You would think.... but perhaps not.  Sometimes peeps are 2-3KC off.

The Sony ICF-2010 had a pretty nice automagic sync detector - would rather seamlessly lock in, and you could select upper or lower sideband by adjusting the tooning knob just a bit.

[WA1GFZ]

The Racal will lock on the strongest signal in the pass band being a FM demodulator. I've seen it lock on a signal 8 KHz away using a 16 khz filter and the bandpass quiet.

[Bacon, WA3WDR]

Thanks for the comment, Stu.  I published a description in the Amrad journal back around 1998 or 1999, so I guess that's one patent that is lost to the public domain.  I wouldn't mind, if only somebody would use the idea... but nobody is doing it that way.

I had been thinking along those lines before I got into DSP.  Then, using the freeware from TAPR and such, it was fairly easy to get the FFT going.  Doing the cross-mixing to separate the sidebands came from a flash of inspiration when I realized that the mysterious 'real' and 'imaginary' were really just another form of I and Q.  Remarkably, the 24-bit 56002 DSP programmed very much like the old 6502 in a Commodore 64, but with three 'lanes' instead of only one.

In my system, the bins were already at baseband, so frequency shift wouldn't apply in this case, but I don't think that people can hear an SSB tuning error as small as 0.1 Hz.  I can hear the 'rolling' of voice harmonic structure when tuning SSB, and with a stable transmitter and a good receiver I can tune it so closely that you can't hear that, and you can't tell it was SSB.  But I think that there is too much dynamic change in typical speech to tell if the tuning is better than 0.1 Hz, or probably even to be quite that close.  I figure I was tuning the SSB within about 0.1 to 0.3 Hz.  Maybe if someone held some vocal sound for a few seconds, it would be possible to hear the harmonic structure rolling, but I don't think you could be sure whether it was harmonic structure error, or selective fading.

I like the idea of dynamic SSB tuning by DSP.  That was another idea that I was considering.  I read an article recently where the author was saying that tuning accuracy of around 0.1 to 0.2 Hz was typical.

I did play with frequency inversion with the FFT by transposing the frequency bins and then converting back to audio.   I could even tune the inversion frequency in bin-sized steps by changing the foldover frequency.  I also stamped the top bins to zero to make a sharp low-pass filter.  That was when I learned that the I and Q bins are in reverse order from each other - I should have been cutting off the top end, but the audio was still there, just a little rough-sounding.  It turned out that the lows were coming from one set of bins, and the highs were coming from the other.  You get some of the audio with only the I channel or only the Q channel, but if both are gone then you get nothing.  I cut off lower in frequency, and a hole appeared in the midrange, because that range was cut out of both.  So then I stamped the lower-address bins to zero in one, and the higher-address bins to zero in the other, and I got the sharp low-pass characteristic I expected.

[WA1GFZ]

Hey Bacon.....I started with the 56303 EVM. Pete SOV modified 56002 code for me. Then I added a few hacks to make it run. It worked quite well. This was before the software pukes made sound cards work. I think we started with Frohne code. That nice EVM is sitting in a box these days.

[Bacon, WA3WDR]

I called that system the Sideband Dodger.

The sound card approach is the way for amateurs to go today, because it is low cost, and everybody has the hardware already.  With today's fast processors, a PC can do a whole lot of DSP.

The 20KHz top end of the sound cards force a decision - if you use it as a single-channel 20 KHz bandwidth IF, it's simpler, but you only get 10 KHz top end with double sideband.  If you use an external IQ demodulator, and use the sound card left and right as dual 20 KHz inputs, you can get 20 KHz audio double sideband.  I think that what is needed is inexensive sync detector hardware that offers I and Q to the soundcard.

The sync detector should be able to lock on carrier or sidebands or both, and it should hold lock if the carrier polarity flips.  That requires a full-wave IF doubler and a Costas loop followed by a 2:1 divider, which could be two phase for I and Q.  There is also a technique that I saw from 1971 from a guy named Rosario Badessa, called a Reciprocating Detector, that is something of a stimulated parametric oscillator that is stimulated by both carrier polarities equally, and functions as the oscillator and the product detector.  However, the RD as designed is only an I-channel device.  It is an excellent and simple sync detector, and if it can be modified to demodulate the Q-channel as well, it would be nice for computer-processed sideband-dodger reception.  A third option would be a DSP-synthesized BFO, or direct DSP demodulation with a virtual BFO.  It's cost and complexity versus performance.

[Ian VK3KRI]

I think that what is needed is inexensive sync detector hardware that offers I and Q to the soundcard.

Ive been thinking about this before, and I don't think you need a sync detectorfor the analog side.  If you detect I&Q with some fixed freq LO and its quadrature, you should be able to offset tune to 'zero beat' in the DSP. (multiply with a software IQ carrier offset by a positive or negative frequency?). That allows you to do all the PLL stuff in software, making for a real simple converter. 

This is currently 126 on the 'to-do one day ' list....
                                                                 Ian VK3KRI