Working CD-4 (software) Demodulator!

[kfbkfb]

Mendemender - Thanks for making and posting the Wonderland needledrop.

If it isn't too much trouble, could you process this same needledrop thru the
(soft fail) CD-4 decoder and post it (I'm interested to hear how the soft fail
decoder sounds on the ping-pong Quad content).


(It's still amazing to me to hear an all software decode of CD-4)

Kirk Bayne

 

[Mendemender]

Here's the processed needle drop. To my ears the hq Sunshine doesn't sound too good. However Cherie Amour sounds better so maybe it's something to do with the musical content rather than record.

Soft fail demodulator
hq demodulator

 

[Brice]

Many thanks for the audio files.

 

[Brice]

OK, well here's a needle drop of the track that the reviewer found so good. Personally I find "ping-pong sound" makes my head hurt You can try it out and form your own opinion.
Wonderland needledrop

Richard, are you reading this? Is there any way to adjust the parameters of the noise/click removal other than just turning it on or off? I normally use DeNoise and ClickRepair and I think a bit more control would be useful.

Hi Mendemender,

Apologies, I haven't looked at this thread for a little while. I was very interested in your results. They sound very similar to what I expect from the two decoders.

The click-pop processing is only applied on the baseband information, not the demodulated signals (although these do have other noise reduction systems applied). In principle it would be possible to tweak all sorts of parameters (it's just software after all!). But how I could present that as part of a coherent interface I worry about.

I think this forum is an excellent place to develop another decoding "recipe". I wonder (for example) if there's room for an intermediate decode; somewhere between the HQ decode and the standard decode. As people use the software and share the results, if there develops a consensus for something like that, I would be very happy to include that in a later issue of Stereo Lab.

Best wishes,

Richard

 

[kfbkfb]

My suggestion is to allow using the soft fail FM demodulator with the HQ/Classic CD-4 Decoder.

(I searched for info about what causes "Sandpaper" CD-4 sound, didn't find much,
maybe another decode mode for severe cases of "Sandpaper" sound, maybe
just use the bass and midrange from the F-B signal and fake the treble with
SBR - Spectral Band Replacement)

Kirk Bayne

 

[Brice]

Hi Kirk,

The sandpaper artefact in CD-4 decode is due to one of three causes:

• Physical damage to the subcarrier information;
• Baseband information getting into the FM decode (they called this "uptalk" in the day);
• Crosstalk between the left and right subcarrier signals.

Rarely, of course, is one of these present and the others aren't; it's usually due to a combination of the three. Of the three, the second is the most significant. Some of the blame has to go to the mastering which (in the USA) opted for too great a modulation level for the baseband signals. I conjecture a reason for that here:

http://pspatialaudio.com/CD4_classic.htm
My suspicion is that, especially towards the end of the quad' era, the record companies "threw the CD-4 customer under the bus" in order to cut a reasonably loud stereo disc.

I'm not quite sure what you mean in your first paragraph but it sounds something like my suggestion that a decoder which was a blend of our soft-fail and the classic might be a worthwhile addition.

Best,

Richard

 

[Mendemender]

My opinion is that a demodulator should recover the music as it’s recorded on the record. I’m not convinced by the approach of soft fail. If I just want to play a record on a turntable to hear some sound then maybe, but if I’m digitizing it once and for all then I want to get it right. I was disappointed when my Marantz demodulator produced bad results. However I suppose that the technology in it will soon be 50 years old so we ought to be able to do better now especially in software. E.g. it should be possible to implement all sorts of filters to get rid of “uptalk”, and I’m sure my AT-OC9ML/II is much better than anything available in the 1970s. If a record is too bad to get a decent transfer then you can always buy a better copy.

I have an uncomfortable feeling about “one-click” processing. I use Brian Davis’ ClickRepair a lot and am used to having various parameters to play with and also being able to carry out processing one step at a time. It would be nice to know what Stereo Lab is actually doing below the surface. Well who knows? Maybe it works fine without user interference but I would feel more comfortable if I had some sort of feedback of how well it’s doing whatever it does. I don’t know how you could implement a Click Repair sort of interface within the framework of Stereo Lab. Maybe some sort of log file and a few more parameters to twiddle before starting the decode?

I also noticed that if I look at the spectrum of music I have recorded with 192000kHz sampling rate then not only is there a good 30kHz signal there but also a quite strong 60kHz signal and a noticeable 90kHz signal. Would it be possible to use these harmonics somehow to improve the demodulation?

Having said all that, I’m very impressed by the present HQ demodulator which already seems to do much better than ancient hardware.

Stuart

 

[Brice]

E.g. it should be possible to implement all sorts of filters to get rid of “uptalk”

Hi Stuart,

Thanks for your comments and I'm glad the software decoder is working well for you.

You are absolutely right about filtering. This is one area in which a DSP based decoder can easily improve upon hardware is by improving the band-defining low-pass and high-pass filters which separate the baseband and FM-band signal paths. Not only may digital filters be designed more precisely and to have a sharper transition region than their analogue counterparts, but they can be made non-causal; allowing a single low-pass filter to be used to separate the baseband signals and simple subtraction to yield a perfect complementary high-pass at the start of the FM signal processing path. This is the approach taken in the new decoder where an eight-pole, non-causal low-pass filter with a turnover at 15kHz is employed to separate the two bands. Although, the cross-modulation of the FM signals from the baseband modulation and between themselves is one of the major failings of the CD-4 system and no filtering, however perfect, can separate the signals once they have been subject to intermodulation.

All the best,

Richard

 

[Mendemender]

Hi Richard,
thanks for the explanation of the filters. There is one thing that I don't understand completely. You say your low pass filter has a turnover of 15kHz. When I look at the JVC description of the CD-4 system, figure 1 shows a line at 15kHz. It's not clear exactly what that is but the data at the bottom of page 3 indicate that may be a -0.2dB point. Whatever, the baseband appears to have components up to approx 17.5 kHz. If you cut off too sharply at 15kHz then you will cut off some of the bits between 15kHz and 17.5 kHz. That might not matter in itself as my ears can't hear those frequencies anyway. However if you produce the 30kHz carrier by subtracting the filtered baseband from the total then won't some of the higher audio frequencies leak through and mess things up? The carrier only extends down to 20kHz so wouldn't say a 17kHz turnover be better? (Or make the frequency a twiddlable parameter for people like me )

All the best
Stuart

 

[Brice]

Hi Stuart,

One of the things I learned in developing the CD-4 decoder is just how limited is the literature. In the end, I learned much more by 'scoping and testing real hardware decoders than I did from standards and articles. You're right, the 15kHz limit is vague in the CD-4 literature (and even in IEC Publication 98). I believe it means the 3dB point of the filter. The only data-point I have for this is the RCA Quadrulator engineering report (JAES MARCH 1977, VOLUME 25, NUMBER 3) where they say of the low-pass filters applied to the baseband signal that they have a "turnover (i.e. -3dB) at 7.5kHz". This was for a half-speed mastering system, so that makes -3dB at 15kHz when played at 33.3 RPM. I think that's what the JVC graph shows too. The droop on the curve and the scale suggests it's a -3dB point.

It's possible therefore that I am "eating" into the baseband information a little with the turnover I chose. But (as you say), I don't believe there's much audible difference.

Your point about the effect upon the decode I am more confident about. During development, I did lots of experimentation with further filtering the FM carriers: all to very little avail. As I said, it's not "raw" baseband signals polluting the FM which causes the problems with CD-4, it's the intermodulation of the baseband signal upon the carriers which is the (very real!) issue.

It's interesting to amplitude demodulate the FM signals (rather than FM demodulate) because that gives a subjective "handle" on the degree that one band of signals (the baseband) is interfering with the other (the FM signals).

I have attached a little mp3 file of a snippet of amplitude-demodulation of a Cat Stevens track (Greatest Hits , known to be a difficult disc). The first part is with the separation filter at 15kHz (8-pole) and the second part with the separation filter set at 20kHz on its lower edge. There's hardly any difference. Everything you hear is crosstalk from the baseband causing amplitude modulation of the FM signals.

It's this severe amplitude modulation which "gets through" into the FM decode no matter how much limiting and amplitude rejection is arranged in the FM decode chain.

The simple reason for this is that the baseband signal on the disc is cut way too hot. As I've said before, I think this is because, as it was realised that CD-4 listeners were obviously going to remain a small minority of the market, the decision was taken to make the disc play reasonably loud in stereo and to heck with the "quadies".

All the best,

Richard

 

[Mendemender]

Hi Richard,

Thanks for that explanation. The Cat Stevens snippet is very interesting. It reminded me of something I read years ago, possible in the Audacity forum. Someone was working on a CD-4 demodulator, possibly written in Nyquist, and claimed it was showing promise. He’d already got a demodulated signal, albeit very distorted and noisy but a few more tweaks and he’d be there. Someone pointed out to him that he was doing an AM demodulation rather than FM/PM, so that was the end of that. The question was why it had produced any results at all. I guess that’s the answer.

Yes, IMD is a real nuisance, generally impossible to filter out. In this case however, the distorting baseband signal is also present. Would it be possible to use that to perform some sort of deconvolution?

All the best
Stuart

 

[kfbkfb]

I was thinking about the original FM demodulators mentioned in post #31
being used in the HQ/Classic CD-4 Decoder, upon further thought, it doesn't
seem like it would be very useful to do that.

Suggested addition to the HQ/Classic CD-4 Decoder:
(F/B) CD-4 Hi-Blend (implemented in the SH-400 and some later model
Technics receivers).

One implementation of CD-4 Hi-Blend is described in the 1976-11 issue of
Audio mag, starting on page 36:
https://www.americanradiohistory.com/Archive-Audio/70s/Audio-1976-11.pdf
^^^
"...the blending action works only on frequencies above about 4 kHz..."

Perhaps listening tests would reveal a better blending frequency or maybe
some sort of dynamic system that changes the blending frequency and
amount of blend based on the difference signal characteristics.

Kirk Bayne

 

[Brice]

Hi Stuart, hi Kirk,

Firstly the idea of blend of the demodulated signals. I know the article by Joseph Giovenco (Kirk you may have sent me this in the past.) Blend does work well and is one of the strategies we use in the fail-safe demodulator.

Stuart, your idea of using the baseband signal to "undo" some of the damage to the subcarriers by baseband modulation is clever and intriguing. We do quite a lot of signal manipulation to the derived carriers after filtering. We calculate the RMS value for the track and adjust so that both channels are matched: this corrects for any imbalance or azimuth error in the cartridge. Then we "limit" (although this has to be done a different way in DSP because we can't go above Nyquist).

If it was possible to discover what was happening to the subcarriers in the presence of large the baseband groove modulations, you're right, we ought to be able to undo that. (Of course, the difficult bit will be finding out the forward transfer-function!) I'll do some experiments are report back.

Best wishes,

Richard

 

[kfbkfb]

Perhaps a check box to (optionally) reduce the LF/LB and RF/RB
channel separation in the HQ/Classic CD-4 Decoder to a fixed
12 dB across the board (and an explanation on the HQ/Classic
web page about why 12 dB is an important number).

This would allow taking advantage of the improved FM demod
in the HQ/Classic Decoder and also mitigate the effect of a
somewhat distorted/noisy difference signal.


ClickRepair and CD-4 discs:
I just skimmed the ClickRepair User Manual, it looks like it
would work to declick a CD-4 disc, it could be cascaded
with the Stereo Lab (possibly with pop/click off) CD-4
Decoders.

Kirk Bayne

 

[proufo]

ClickRepair and CD-4 discs:
I just skimmed the ClickRepair User Manual, it looks like it
would work to declick a CD-4 disc, it could be cascaded
with the Stereo Lab (possibly with pop/click off) CD-4
Decoders.

Don't know how this demodulating works, but at one point there must be a regular wav file of the FMd tracks.

If it doesn't, it would be great if made available, not just to apply ClickRepair and the such but also for those who want to do the additions outside StereoLab, for example using the parameters derived from a separation adjustment disk.

 

[Brice]

Hi Stuart, Kirk and Pablo,

I tried using the baseband modulation to post-process the sub-carriers. So far, the test has established that there's no simple technique.

I derived an envelope signal from the baseband and used this to modulate the carriers inversely. The results, in terms of the carrier envelope was much improved. That's to say, it looked much more constant-amplitude. I was encouraged!

However, FM demodulation of these "improved" carrier signals revealed that the demodulated audio was even more distorted than the un-corrected signals.

This is interesting because I had always thought that poor demodulation results simply came from some remnants of amplitude modulation getting past the limiters and demodulator (which is designed to be amplitude insensitive). I was obviously wrong.

I think what must be happening is that stylus tracing distortion, due to heavily modulated baseband modulation, is actually causing angle modulation of the carrier. This was recognised as a problem late in the quad' era (RCA Quadrulator engineering report JAES MARCH 1977, VOLUME 25, NUMBER 3). This picture explains the problem well:

So, I switched to the idea that maybe I could add (or subtract) some of the baseband from the demodulated signal on the basis that, if this "uptalk" was genuine angle modulation then it would add to the real FM modulation. But, of course, the problem here is that, you can't "add" FM modulation and expect it to simply crosstalk as AM modulation would. (That is the essence with the problem of crosstalk between the L/R carriers.)

For the moment I don't see a way around this. It is important to stress again that this "uptalk" problem came about by abusing the system anyway. I used the Cat Stevens record for my tests. It really is mastered with little regard for CD-4 customers.

Richard

 

[Mendemender]

Hi Richard,

That was quick. I was expecting you to be playing around for the next couple of weeks

if it really is the problem you suggest then as illustrated in your picture, the phase error would depend on the "gradient" of the baseband. How about differentiating the baseband and applying a phase correction based on the result? (but will that problem occur with a line-contact stylus?)

Otherwise, if you could determine a transfer function then you might be able to do a deconvolution (lots of Fourier transforms ). Maybe there is a CD-4 test record somewhere with something simple like a pure sine wave on it. Then you could look and see what appears on the carrier. For example a sine wave FL should produce the same signals in the left baseband and carrier. Any differences should give you an idea what's going on.

I wonder why they "abused the system". After all, if it's stereo that's important you could just produce a stereo record. Issuing stereo as a CD-4 record which only works in stereo would seem to be a bit pointless. Did they increase the modulation of the carrier as well. If they only boosted the baseband then that would mess up hardware decoders where the levels had been adjusted to provide max F/B separation.

All the best
Stuart

 

[kfbkfb]

IMHO, the ongoing pressure for ever higher CD-4 baseband levels was partly instigated
by the Matrix Quad (SQ, QS) proponents (trying to discredit CD-4).

IIRC, many CD-4 related articles in the (USA) consumer audio related publications
made it a point to mention that CD-4 baseband levels were being increased by using
new mastering systems, new cutting stylus shapes etc.

The article link in post #132, on page 38, states that one CD-4 title has a baseband
level 2dB higher than the Stereo release of the title.

Kirk Bayne

 

[proufo]

Could a micro ridge stylus be useful?

Could half-speed playing be useful?

Ditto tangential playback with perfect alignment of the stylus tip relative to the groove in the left-right direction. It has been reported that styli not always are perfectly aligned to the cantilever.

 

[Brice]

Hi All,

I'm sure Kirk is right, the pressure to raise the level of CD-4 records came from goading from the rival systems' marketing departments and from disappointment from stereo users at how "quiet" the records were. Remember, the whole (and flawed!) business model for quadraphonic was for "single inventory" compatible stereo/quad discs.

I've attached two frequency analyses of a US vs. a Japanese CD-4 recording. See how the baseband information is 12 dB higher, relative to the carrier, in US recording (cat_spectrum) compared with the Japanese (Gaye_spectrum).

Stuart, the LEVEL of the carrier isn't increased relative to the baseband. To do so, would have set fire to the cutting-head coils! This is FM and the carrier is much more heavily deviated and that's obvious from the spectrums too. See how broad are the sidebands around the carrier in cat_spectrum. That of course makes everything worse too because the sidebands are sploshing into the baseband. Yet another source of distortion as these are lopped off in the carrier acceptor filters. The output of any electrodynamic cartridge is always the time-differential of the groove modulation.

Hi Pablo, yes, I guess the narrower the radius of the reading stylus, the better. I haven't experimented with micro-ridge for CD-4 records but I have heard good reports for people who use them. Half-speed replay is absolutely a help if you have a deck which runs at 16.66 RPM. See:

http://pspatialaudio.com/half_speed.htm
Stereo Lab will automatically work at half-speed and do the necessary changes to double the playback speed on decode. It's engaged by a check-box in the PHONO Preferences.

I have got some instructions for modifying a t'table to do 16.666 RPM here:

http://pspatialaudio.com/turntable_mods.htm
There is one consideration with half-speed needle-drops that spurious low-frequency signals caused by motor or bearing noise or due to disc imperfections (perhaps exaggerated by a poorly chosen and controlled resonance of pick-up arm mass and cartridge compliance) will be frequency doubled along with the legitimate audio modulation. See the half_speed.htm page for information about this.

Richard