Lou Dorren: A new CD-4 Demodulator!!! [ARCHIVE]

[quadtrade]

This gig was a dream we all missed.

 

[Owen Smith]

This gig was a dream we all missed.

It would be great if it happens, and I'd buy one (to lend to a friend), but I'm realistic about the chances of this becoming reality.

 

[Q8]

Is the design for this completed? If he gets to a point where he is sure it will never happen, I would hope he would release the design to the quad world. This probbly uses commercially available parts except for the PCB(s) and case, we could do our own crowdfunding type thing to make PCBs for this if it came to that. Then it could be a build your own demod type thing...

 

[FM_Quad]

Is the design for this completed?

The schematic is in this thread. I had heard the Bill of Materials was posted earlier, but I haven't seen it. About the only thing I don't know about and couldn't go out and buy is the filters. I think they are off the shelf, but I don't know who makes them. Also his arrangement for the cartridges/stylus, although it could probably be re-negotiated by someone. The PCB and enclosure would take some work but certainly aren't impossible.

OTOH, if I went to that much trouble I'd go high end digital with it (which I still may do if I can find the time).

 

[Q8]

I thought it was hard and/or impossible to digitize the CD-4 process.

 

[FM_Quad]

I thought it was hard and/or impossible to digitize the CD-4 process.

I'm almost certain it isn't impossible, and most of it should not be hard. The filters are trivial, and work better in digital. I want to test the FM demod a bit, but I don't think that is too hard. The ANRS could be the most challenging, but the biggest challenge even there is knowing the exact specifications. There were a number of patents and financial interests associated with CD-4 and so the details weren't widely disseminated. The patents have certainly all expired, but so have most of the people who knew the details. Lou is the only person I know who has that information, and he isn't terribly accessible. I have collected everything I can, and when I get a bit of time, I will see how far I can take it.

The basic approach is to digitize at 196 KHz so it is significantly oversampled, using a pair of 24 bit ADC so digital artifacts don't interfere with the process. Then a low pass filter for F+R on each channel and a high pass filter for F-R on each channel. Then mix the L-R carriers with 30 KHz in quadrature to shift it to 0 IF. This gets low pass filtered and a second low pass filter that differentiates which feeds a digital discriminator (From Rick Lyon's book) to decode it. All of that is straightforward.

Then there has to be some frequency shaping. RIAA de-emphesis in F+R, which should be straightforward. FM/PM filtering in the decoded carrier, which isn't real clearly defined. And finally the ANRS, which is a combination of filtering and gain control. It should actually be easier in digital, because there is nothing that says this decoder has to put out sounds instantly when it receives them. If the sound came out 1 second later, the listener would not really know the difference. But a few ms of delay might allow the logic time to "look ahead" to make gain control decisions that would otherwise be harder with digital. The only requirement is that all signal path delays be the same, beyond the delay put into F+R when it was recorded that has to be matched. That too is another area that is easy to implement, but not as well documented as I would like.

At this point, the F+R and F-R signals are ready for final combining (also trivial). Add them together gives 2F +R -R which is simply twice the amplitude of Front with rear cancelling. Subtract them gives F - F +2R so the front cancels. After that it goes to four DACs for analog output or SDIF or something for digital output.

 

[ndiamone]

Trouble with the 192KC sample rate (double 96) is - at the upper 45KC limit of the carrier wave you might only get 5 or 6 samples of the ENTIRE WAVE even though at the low end of 30KC you might get a couple more - maybe 7 or 8 samples of the entire wave. Certainly nothing you can effectively demodulate from.

The only thing that MIGHT work is if somebody has access to Jamie Howarth's still-in-effect patents for being able to retrieve bias frequencies off magnetic audio and video recording media - and amplifying them enough to use in driving a resolver a.k.a. sync apparatus in order to correct for the miniscule and continuously-variable speed variations of analog recordings thousands of times a second or hundreds of thousands of times a second.

Continuously correcting the speed in this manner through his patented process, described in detail at www.plangentprocesses.com removes many of the effects of the intermodulation and beat frequency problems as well as much of the other grunge that makes its' way into analog recordings.

Other than by using the CD-4 experiments from the early 2000's which tried to sample full bandwidth PCM and PWM in the megacycle range normally used for video - that's the only way I can see to get an acceptable pre-demodulated CD-4 signal captured to digital since the CD-4 and UD-4 carrier waves are in the same range as a great many magnetic bias frequencies from the middle third of the 20th Century.

But the reason those all failed is mainly because of the process being a data hog. Think about it: For a 45KC sine wave to be adequately captured, it needs at least ten samples per sine wave - and most physics books will tell you a hundred times over your highest frequency to get an accurate capture.

So 45KC times 10 is 450KC and times a hundred is 4500KC or 4.5MC. And THAT's just for CAPTURE.

Now try dealing with editing and further processing on uncompessed FILES that take up 4.5MC per SECOND and you won't wonder why the few of us that were able to do so - juryrigged half-inch instrumentation recorders to be able to capture the 30-45KC carrier wave intact - so that the resulting recording - sans reverse RIAA equalization - could then be re-demodulated through a standard CD-4 demodulator without wearing out the record.

So like I said - if I were the guys working on bringing this to fruition in 2014 - I would start having my engineers have conversations with Jamie Howarth to see how his research into capturing and using bias tones applies to CD-4.

 

[FM_Quad]

That really isn't the case. The science behind this says that if you sample at > 2 x the highest frequency component, you can accurately and completely reconstruct the waveform. There are practical considerations beyond that theoretical limit, such as how rapidly filter attenuation can increase to prevent unwanted signals (noise) from sampling and aliasing down to the range of interest, so we generally try for 2.5 or 3 time. But with the highest frequency being 45 KHz, 192 KHz provides plenty of margin. There is also the reality that the subcarrier is frequency/phase modulated and that produces an infinite set of sidebands on either side of the 30 KHz carrier. However, the modulation index for higher frequencies in this case is quite low, so almost all the energy is contained in the carrier and the first pair if sidebands. Also the mechanical physics of the situation insures that sidebands above 50 or 60 KHz simply weren't recorded to begin with, and wouldn't be picked up by the cartridge if they were. The situation below the carrier is even more stringent. Any sidebands extending below 15 KHz had to be thoroughly removed before the master was cut, or they would have shown up as noise in the F + R channels.

It is true that it is hard to make a good discriminator without over sampling. Simple models like a sample rate of 10 x the highest frequency, which would be around 450 KHz. The limiting factor is the ability to create a good differentiator. That can be done very nicely with FIR filters, though. Even if that were to be a problem, 3x interpolation of a 192 KHz data stream would accomplish the purpose nicely. There is no information above 45 KHz so interpolation simply gives more samples to work with, as discriminators (especially simple ones) like a wide difference between their pass band and their stop band.

It is true that all of this is computationally intensive. The standard measure is the MAC (multiply accumulate) A filter for this application (of which there have to be half a dozen, times two channels) might take 100 or 200 MACs and whatever sampling rate is used. At 192 KHz That is about 40 MMACS (million MACS). There are DSP chips that can to several hundred MMACS available in the $10 to $30 price range. Also FPGAs have hardware that can do 200 MMACS or more, and have a dozen or two of them on an IC that also is in the $10 to $30 price range. The number of bits involved has an impact here as well, as many of the devices are designed for 16 to 18 bit operation and I wouldn't attempt something like this with less than 24 bits in order to maintain fidelity. However there are 32 bit DSPs. Also 4 MACs in an FPGA can be combined to provide 32 bit precision that way. According to my calculations that would require about 20 MACs, which would require something like a Xilinx XC6SLX16 at about $25.

 

[ndiamone]

That really isn't the case. The science behind this says that if you sample at > 2 x the highest frequency component, you can accurately and completely reconstruct the waveform.

If that were the case, then why do all the proprietary-format AEI audio CD's (sampled at I think 12 bit and 37.something KHz to get four hours of stereo music on one CD) all sound crappy?

Music (supposedly) only goes to 15KHz like on FM Stereo radio - so a 37.whatever sample rate and 12 bit word length should have been more than enough for commercially-produced CD's as well according to your math.

And if THAT was the case then why is there SACD's and DVD ProAudio and Blu-Ray Multichannel High Resolution and on and on and on if twice the same rate of the highest frequency was ``good enough''?

 

[Owen Smith]

If that were the case, then why do all the proprietary-format AEI audio CD's (sampled at I think 12 bit and 37.something KHz to get four hours of stereo music on one CD) all sound crappy?

Music (supposedly) only goes to 15KHz like on FM Stereo radio - so a 37.whatever sample rate and 12 bit word length should have been more than enough for commercially-produced CD's as well according to your math.

And if THAT was the case then why is there SACD's and DVD ProAudio and Blu-Ray Multichannel High Resolution and on and on and on if twice the same rate of the highest frequency was ``good enough''?

I've got a 24/96 bd audio which sounds crap. It's a bad mastering job. This may be what ruins the 37Khz 12 bit CDs, which as you say is as good as FM radio so could sound OK.

As to why people thought we needed SACD and DVD-A etc, it's just a case of ill informed people assuming that something with a higher number is better. Many of them do sound better, but that's due to a superior mastering job and not the bit rate and depth. CDs are usually ruined by hot mastering.

 

[FM_Quad]

If that were the case, then why do all the proprietary-format AEI audio CD's (sampled at I think 12 bit and 37.something KHz to get four hours of stereo music on one CD) all sound crappy?

Music (supposedly) only goes to 15KHz like on FM Stereo radio - so a 37.whatever sample rate and 12 bit word length should have been more than enough for commercially-produced CD's as well according to your math.

And if THAT was the case then why is there SACD's and DVD ProAudio and Blu-Ray Multichannel High Resolution and on and on and on if twice the same rate of the highest frequency was ``good enough''?

A number of good questions. First, the range considered audio is 20 to 20,000 Hz. Most of us can't hear all the way to 20,000, but a few can. I can still hear to at least 16,000. Second, 37KHz isn't that bad of a sampling rate. FM stereo broadcasting has a subcarrier at 38 KHz, and we've enjoyed listening to that most of our lives.

The big if in you question is the 12 bits. 12 bits is enough to capture about 72 dB of dynamic range. That may sound like a lot, but it isn't. First, unless a lot of compression and limiting is done, there needs to be at least 10 dB of headroom for peaks. That means that your normal 0 Vu recording level needs to be at least 10 dB below the peak capability of your medium (in this case digital word). That has reduced your dynamic range to 62 dB (theoretical, real world issues are going to take at least 2 dB away from that, and more likely 5 or 6). Some of the early digital music synthesizers used 12 bit. Normal sounds weren't too bad, by decay (particularly things like a bell fading out) sounded awful. It gives very little room for manipulation. Also the human ear can hear a range of at least 90 dB.

Things get more interesting when you look at the details of sound, though. 60 dB SNR was the FCC limit for FM broadcasting, and a lot of analog consumer equipment only came in around 55 dB (some closer to 45 dB). But that it total noise energy spread across the entire audio specrum. The human ear is able to hear discrete frequencies well below the level of broadband noise, so just because the analog SNR is only 60 dB doesn't mean that its OK to limit the digital dynamic range to 72 dB. There is also another aspect that a lot of people ignore. The number of bits in digital leads to a condition called quantization error, with an artifact known as quantization noise. The bottom end of the dynamic range is a hard cut off at the least significant bit. Not only that, but the handling of the LSB is generally not perfect. The result is a low level of intermodulation distortion. IM has a very harsh, metallic sound that is quite noticeable, so it needs to be kept well below both the level of signal being recorded and well below the wideband noise floor, or it will detract noticeably from the content.

So the number one problem with the bad quality sources you cited has nothing to do with sampling frequency, and everything to do with number of bits. I do not consider anything to be a high quality digital system that uses less than 24 bits per sample (per channel). While I generally enjoy listening to CDs that are recorded using 16 bits, I know people who can't stand to listen to them, because they can hear the IM and it sounds harsh, metallic and grating. Of course pragmatically, manufacturers are trying to design systems that use as few bits (both in data size and in sample rate) as possible, because they can store more information, and transfer it faster over a given medium (such as the internet). But this leads to compromises. We had compromises in the analog world, also. Records varied quite a bit in noise performance, for instance, depending on the exact vinyl formulation. Reel to reel tape varied depending on the tape used, the speed, the number of tracks and the width of the tape (both of which determine the width of the track).

Another factor that impacts a lot of digital recording is compression. MP3 (the most common one) is not a lossless compression technique. It creates a mathematical model of tones (rather than sounds) because such a model can convey most of the information with far less data than raw samples. Then it tries to reproduce the original sound from that model, on playback. When done well it can sound fairly good. But I've heard some low bandwidth MP3 signals that were pretty harsh (part of a DVB signal). Any high quality music recording process should either use no compression (like PCM) or should use only lossless compression.

As far as the sampling rate is concerned. The theory remains as stated. If you want accurate response to 15 KHz then you have to sample at 30 KHz or above. IF you want accurate response to 20 KHz then you have to sample at 40 KHz or above. Note that standard CDs do that (barely). The CD-4 standard actually does that, as the 30 KHz subcarrier is mathematically equivalent to sampling at 30 KHz. The theroretical maximum audio frequency that can be reproduced through a CD-4 record is 15 KHz.

But. There are real world concerns here as well. If I am sampling at 30 KHz and I encounter a audio tone (probably a harmonic of something, or possibly part of a wideband signal such as a cymbal crash--or even a snare drum brush stroke)--Let's say it has a frequency of 15.4 KHz. It will be digitized as if it was 14.6 KHz. Besides being wrong, that has no harmonic relationship to the original so it will sound rather harsh and foreign. So, if anything can possibly be above that frequency (and this has nothing to do with whether the human hear could have heard it or whether it would be desirable to reproduce correctly) it must be filtered out before the digitization is done. The problem is that you can't make a filter that passes 14,999 with no loss and reduces 15,001 to the level of inaudible. We have technology today that can do fairly well, but not that good. The reason CDs use 44.1 is that the filter has to cut off at 22.05 and it isn't that difficult to pass 20,000 and reject 22050. But the point is that the sampling rate has to be sufficiently above what is needed to allow a filter to be designed that has room for its transition band.

There is a second issue that has to be dealt with, also. If you want the mathematical background, do some research on the term sin(x)/x. It turns out that the sampling process doesn't capture the amplitude completely accurately. There is a frequency dependent factor that gets worse, the closer you get to the sampling frequency. There are ways of compensating for this with filters, but that has side effects, too (like changing the phase). The variation is fairly small, and since it is worst near the top edge of the frequencies, where it is probably least noticeable, is is often ignored. However, these are the two reasons why most professional audio recording is done at 96 KHz. It give a lot more room for the filter transition band, resulting in a simpler filter and less filter artifacts. It also means that the actual audio is in a portion of the spectrum where sin(x)/x errors are very small (like .02 dB). Even if the end product is going to be put on CD (at 44.1 KHz) or converted to MP3, it is advantageous to sample at 96 KHz, do whatever filtering and other processing is necessary in the digital domain and then do sample rate conversion to the lower rate, with everything as carefully prepared digitally as possible, so the end product is as close to the theoretical ideal.

Side note: That is also why high end CD players use more than 16 bit DACs and sometime sample rates higher than 44.1 KHz. It allows them to take the original (which hopefully was generated under close to ideal conditions) and convert it back to analog under conditions closer to ideal. It will never sound better than what is theoretically possible for 16 bit PCM and 44.1 KHz, but it will approach that ideal more closely. How much? maybe only 1 or 2 dB, but it will make a small difference.

Now for CD-4 decoding. If the sample rate is 192 KHz, the theoretical bandwidth is 96 KHz. The record has nothing on it above about 50 KHz, so there is plenty of margin to do the filtering, etc. right. If it is done with 24 bits the theoretical dynamic range is 144 dB which has plenty of margin for headroom and quantization error and still is well beyond the dynamic range of human hearing. (There's even enough room for doing the RIAA de-emphesis digitally). 24 bit ADCs don't actually give 144 dB of dynamic range. The requirements on the analog hardware doing the digitizing are just too unreasonable. The best do about 110 to 120 dB of dynamic range, but that is still 18 - 20 bits worth (theoretical), so it is worth the effort.

I hope that helps. A lot of listeners come from more of a music background than an engineering background. While we all pick up numbers and "facts" and have to have some sense of what is better and worse, most lack the technical background to attribute what they don't like in what they hear to the correct cause. For instance, digital is not BAD. It is just that digital is often done badLY. Companies exist to make money and they make money by making compromises--always have. Some compromises are more acceptable than others. Fortunately, I am a practicing engineer and software developer who happens to have an interest in music. I earned a minor in music in college. So I'm in the unique position of having "one leg on each side of the fence".

 

[FM_Quad]

I've finally gotten a little time to look at this. I picked up an M-Audio ProFire 610 off E-Bay as a capture device for some early tests. I used it to rip some CD-4 tracks to experiment with. I have a Technics SL-2500 with AT14 cartridge and aftermarket Shibata stylus. (Wish I had a 451C).

I was a poor college student about the peak of the CD-4 run, and after graduation picked up the AT14 with what meager funds I had. My recent efforts started with my desire to digitize my LP collection. I didn't even have a phono pre-amp any more to do it, but I did have a QC-04 demodulator that I had picked up as salvage after quad went bust, which got me thinking in that direction. The QC-04 did OK for stereo, but it badly distorted on the sub-carrier, and I can't figure out why. That was when I found this group and signed up for Lou's new demodulator. I also realized that I didn't actually own any CD-4 records, due to my limited finances back then and the subsequent lack of releases shortly afterward, so I bought 10 or 20 in good condition on E-Bay including a test record. (I wish I had known about Lou's test record offer in time. It is better.)

So that's my background for this. I think I mentioned previously that while in college I was chief engineer of the local FM station and Lou worked with us as I was building a quad sub-carrier generator for the station. We actually never broadcast 4 channel material through it, but did use it for stereo for a few years, and thanks to him had 451C cartridges in the studios.

OK, back to 2014. I started playing with some of the ripped tracks. I wrote some test code for a PC. It isn't real time. It isn't optimized. Nor does it have any of the sub-carrier equalization or ANRS in it at this point. What it does have is the bandpass filtering for the sub-carrier and the FM demodulation and low pass filtering. I wanted to see if it would work at all. Given the pieces that are missing, it isn't bad. It's a start. Now to find more time to sort out the other parts.

If anyone is interested in doing something similar, I put a zip file out on a server with a couple of files in it. It is located at www.esrdsn.com/Samples.zip. Inside are two files. LeftFront.wav is what I ripped off of the LP. F_R_LeftFront.wave is the demodulated sub-carrier from my test program. Both are 16 bit wav files sampled at 192 KHz. I captured them in 24 bit, but couldn't figure out how to get Audacity to save them that way.

Windows media player is lousy at playing them. It is clever enough to know that they are sampled at 192 KHz, but not clever enough to play them at that speed (or the hardware may not be capable). It downsamples them, but without filtering, so the sub-carrier aliases into the audible range and makes the whole thing sound nasty. The current version of Audacity can play them just fine. It also has an RIAA filter that can be applied. If you apply that to the LeftFront.wav file, you should get a fairly accurate stereo (F+R) reproduction of the track. (It appears to me that they are using FIR techniques for the RIAA, which is not correct and will cause phase problems.) Audacity also lets you see the sub-carrier in the waveform (caution, RIAA will wipe most of it out). The F_R_LeftFront.wav file is the decoded F - R, which in this case should be identical (since the tone is located at left front) except for the fact that it needs more processing. I also ripped the "Look What They've Done to My Song" track from New Seekers. I could supply it if anyone wants to mess with it. It is 155 MB long, though, so not for the faint-hearted. I haven't played with it yet. I don't even know if my program in present form can allocate enough memory to process it.

Another useful tool for working with these files is Scope-FIR. It isn't as fast or simple to use, but has more powerful capabilities for processing and graphing.

Anyone who wants to look at the C code and play with it, it also welcome. My end goal is to produce a demodulator as a stand alone device and sell copies to anyone that wants it, but this is my starting point for experimenting. When I get a program that does a credible job of decoding, I will turn my attention to how to implement it in hardware.

 

[ArmyOfQuad]

My line of thinking used to be like ndiamone's, that 96kHz would certainly not get the carrier that well, and 192 would be questionable since you'd not get very many samples per period. But, I'm familiar enough with digital audio that I know there's science behind the whole thing about being able to get the frequency range up to half of the sampling rate. So, as I mentioned in another thread, I decided....eh, lets try it. So to see if a digital file could be demodulated, I recorded a few cd4 tracks, including a test record, in stereo. I ran a reverse riaa eq on them. I fiddled with the levels to get it in the range that the demodulator was happy with. And, I got as good of a demodulation as I've been getting lately going straight off the record....which is not perfect, and some sandpaper or lots of distortion on difficult tracks, but....the digitization to 96kHz didn't degrade the process any. So, if anyone doubted the science, I think that proves that 96kHz cd4 is possible. Although I agree 192 would probably be better if possible, I just don't have the ability to do 192 at this time.

 

[ndiamone]

All I know is that if the carrier wave would have been able to be reproduced accurately at even 192 nevermind 96 - then Jamie Howarth wouldn't have needed to start over from scratch and build whole hardware apparati including writing his own codec and other software to capture the bias tone from tape in order to erase the miniscule speed variations endemic to analog recordings.

http://www.plangentprocesses.com

Granted the bias tones are not modulated and therefore do not need to have music recovered from them - but the principle is the same. He had to start over, develop special bias-reading heads that wouldn't clog - and then develop hardware based on the old film-sync resolver used in dual-system film production - so I think the same process would work well with CD-4.

 

[FM_Quad]

Thanks for the reply and affirmation. No question, a 96 KHz sample rate fed back to an analog decode should be indistinguishable, provided it was done properly. Be aware though that 96 KHz sampling rate doesn't guarantee that the filter cuts off near or right at 48 KHz. A lot of 96 KHz audio gear cuts off around 30 KHz--far enough from 20 KHz so the filter won't impact the sound (phase is the most likely culprit of ignoring this one) and far enough from 48 KHz that the filter doesn't have to be too complex. Obviously that is going to wipe out part of the sub-carrier. Often 192 KHz systems cut off around 75 KHz.

FM demodulation is one of the more challenging conversions to do, and more so in digital, so a higher sampling rate makes it easier to get right. However, even if the demodulator needed a higher sampling rate it would work just as well to interpolate (up-sample) 96 KHz or 192 KHz to whatever was needed as to digitize it at a higher rate. There is no energy above somewhere between 45 and 50 KHz on a CD-4, so there is nothing to get lost.

 

[FM_Quad]

All I know is that if the carrier wave would have been able to be reproduced accurately at even 192 nevermind 96 - then Jamie Howarth wouldn't have needed to start over from scratch and build whole hardware apparati including writing his own codec and other software to capture the bias tone from tape in order to erase the miniscule speed variations endemic to analog recordings.

http://www.plangentprocesses.com

Granted the bias tones are not modulated and therefore do not need to have music recovered from them - but the principle is the same. He had to start over, develop special bias-reading heads that wouldn't clog - and then develop hardware based on the old film-sync resolver used in dual-system film production - so I think the same process would work well with CD-4.

I don't know anything about Jamie Howarth, and the link did not provide enough technical information to determine what he was doing. For what it is worth, the bias frequency on a magnetic tape is NOT a form of modulation. It is simply a trick to eliminate zero crossing distortion caused by the hysteresis that is inherently present in any magnetic material. It is probably true that some of that energy is stored on the tape, but very little. It is a clever idea to try to recover that information to use as a timing reference. It would be very low amplitude, much lower than the sub-carrier on a CD-4. I haven't worked on tape recorders in a number of years, so I have forgotten the range of frequencies used for biasing. I can guarantee you that he didn't have to sample at more than a little above twice the bias frequency in question, to do the phase locking he was attempting. Given the nature of what he is doing, he probably designed it to accommodate a wide range of media. The other possibility is that he didn't really understand it and over-engineered it, just to be "safe". I see a lot of that.

 

[Owen Smith]

When you're building a low volume special bit of kit, over engineering it "just to be sure" is not necessarily wrong. It's better than ending up with something that doesn't work. The downsides are it costs more, and misinformation can be perpetuated.

 

[ndiamone]

Often 192 KHz systems cut off around 75 KHz.

Well maybe that's why then. Everytime I try and demod off of 96 I get the same sandpapery sound as AOQ was getting ``on difficult tracks'' except I was getting it on ALL tracks.

Conversely when I taped the CD-4 LP onto my Viking half inch instrumentation recorder that's supposedly flat out to 100K - my demods from there are pure and clean - even with my thrashed copy of Elvis: Aloha From Hawaii that I use for my ``sandpaper'' test - and even that came out moderately good.

And everytime I tape into the computer at 192/32 - same story. Nice and clean demods with very little IM.

I have forgotten the range of frequencies used for biasing.

The lowest frequencies from semi-pro recorders back in the 50's could be as low as 30KHz - the same as the center CD-4 frequency. However the top frequencies used surpass 100K on some professional NASA spec analog recorders.

He's patented his process. All the technical papers are online and also at AES for review.

 

[Owen Smith]

Well maybe that's why then. Everytime I try and demod off of 96 I get the same sandpapery sound as AOQ was getting ``on difficult tracks'' except I was getting it on ALL tracks.

Conversely when I taped the CD-4 LP onto my Viking half inch instrumentation recorder that's supposedly flat out to 100K - my demods from there are pure and clean - even with my thrashed copy of Elvis: Aloha From Hawaii that I use for my ``sandpaper'' test - and even that came out moderately good.

And everytime I tape into the computer at 192/32 - same story. Nice and clean demods with very little IM.The lowest frequencies from semi-pro recorders back in the 50's could be as low as 30KHz - the same as the center CD-4 frequency. However the top frequencies used surpass 100K on some professional NASA spec analog recorders.

He's patented his process. All the technical papers are online and also at AES for review.

I can't see why you'd need 32 bit, 192/24 should be fine. Unless you mean 32 bit float, which makes a lot of sense as it gives 24 bits of resolution with 8 bits exponent so setting levels is much easier.

 

[FM_Quad]

When you're building a low volume special bit of kit, over engineering it "just to be sure" is not necessarily wrong. It's better than ending up with something that doesn't work. The downsides are it costs more, and misinformation can be perpetuated.

Very True, and in the case in point, the exact nature of usage is somewhat unknown and varied, making it even more justified.

Well maybe that's why then. Everytime I try and demod off of 96 I get the same sandpapery sound as AOQ was getting ``on difficult tracks'' except I was getting it on ALL tracks.

Yes, if the filtering was down around 30 - 40 KHz where it often is, that would have that effect. And yes, it will show up on all channels, because LF, for instance is 1/2 (F + R) + (F - R) and LR is 1/2 (F + R) - (F - R), so the subcarrier contributes to each channel. If you could somehow kill carrier detect so F - R got muted and it was playing back in stereo, it would sound just fine. One way to do that is to not demodulate it. However don't try to play it back through Windows Media player, as it down samples but doesn't filter, so the carrier aliases to about 18 KHz and produces some really nasty artifacts. Audacity will handle it properly.

I can't see why you'd need 32 bit, 192/24 should be fine. Unless you mean 32 bit float, which makes a lot of sense as it gives 24 bits of resolution with 8 bits exponent so setting levels is much easier.

All very true. 32 bit float is no better than 24 bit integer. 32 bit integer is pretty unachievable. The best A to D converters I am aware of are 24 bit, and even then the actual digitization quality is closer to 20 - 22 bits, which is par for the course on cutting edge converters of any kind. All the last 2 - 4 bits do is give a little more dynamic range (soft passages have more bits of resolution), but to actually pick something out (such as filter other stuff away and leave it) down in those low bits when the overall signal fills the whole range, won't produce anything worthwhile. But 24 bits is very adequate for any audio purpose. Even doing RIAA digitally, which alters the dynamic range can be done very adequately with 24 bits.

The lowest frequencies from semi-pro recorders back in the 50's could be as low as 30KHz - the same as the center CD-4 frequency. However the top frequencies used surpass 100K on some professional NASA spec analog recorders.

I sort of remember working with bias frequencies in the 30 to 60 KHz range (again its been many years), but certainly if he was making a solution that would work with bias frequencies over 100 KHz, then clearly 192 KHz sampling rate would not have been adequate. Even with perfect filters (which don't exist), it would alias, and with realistic filters, it would definitely miss the boat.

Speaking of A to D converters and filtering, the best current converters use the Delta Sigma approach which is a low bit, highly oversampled method. One of the advantages of it is that the analog filtering only has to prevent aliasing at the oversampling frequency, which is up in the MHz region, so has no adverse impact on audio frequencies. The audio based anti aliasing filtering can be done digitally, where techniques readily exist that do not cause phase distortion. It is also the only way current technology can get to 24 bits.

Ironically, I ran across some guy's web site that is selling supposedly top of the line audio A/D. He has a few "non-technical" hangups--as apparently some of his followers/customers do, too. First, he builds them on wooden boards. Somehow that imparts some magical quality. I'm not quite sure what. The shielding capability of wood is pretty minimal! He also uses 16 bit converters, because he has fabricated some flaw with Delta Sigma converters and 16 bits is the highest resolution he can buy in a successive approximation converter. Its people like that that feed the constant hysteria of mis-information wafting through audio circles. I'd be happy to put his 16 bit $2000+ wonder up against my M-610 (< $200 on E-Bay) any time. But I digress.