Rerurn to Romy the Cat's Site

I posted this request at the Yahoo FM group, the place reportedly owned by the Reichsführer-SS Bob Fitzgerald, I have mention that Moron-wanna-be-in-charge before. Thankfully that cretin does not show up there and the air in that bulletin blog is free from his stench. Other folks in there looks like fine. Anyhow, here are some of the  responses that I got there on subject.

//*****************************************************************

The way I understand it, the number of bits mainly correlates to the maximum signal to noise ratio...or noise floor in a linear digital system.  The sampling rate defines the absolute frequency response limit.

16 bits should be capable of a noise floor that's at least 90db down from 100% modulation (75khz of deviation).

I think it's unlikely that you can get a noise floor in the real world of FM reception that would exceed 90db.  Most stations don't have their noise floor that far down anyway....especially not in stereo...although some (but very few) are close.

So, (IMHO), I'd say 16 bits should be enough resolution.  If more bits are easily available, that's fine, but 16 should be at or beyond the rest of the signal paths noise floor.

A 44khz sampling rate would limit you to a maximum frequency response of about 20khz.  Since FM is limited to about 15khz, a 44khz sampling rate would be enough.  Moving the sampling rate higher does allow for less abrupt filtering however, and some folks think that sounds better.  But all the frequencies transmitted will be reproduced easily with a 44khz sampling rate.

Hope this helps a little,

Dave O.

//*****************************************************************

A few years ago I needed a way to record FM broadcast in order to do A/B comparisons of small adjustments I was making to the processor of an FM public radio station. In order to do this accurately, I needed a recorder whose reproduced sound was very close (if not identical) to the original being recorded. Of course, I was at the "mercy" of the A/D and D/A converters in the recorder, as well as its limitations of bits and sample rate.

So there were several independent contributing variables. Suffice it to say that, in my case, I was unable to come up with a 16bit 48kHz recording that exactly preserved the definition and sound to my ears. For example I tried a Panasonic SV-3800 DAT recorder at 16bits, 48kHz. (Close, but not quite exact.)

I did hit the "jack pot" with an old Alesis ADAT Type II XT20 20-bit Digital Audio Recorder. With it running at 20-bits, 96kHz, I don't think I could tell the difference between the FM source and the digital recording. It was extremely close, and good enough to call "identical" for my purposes. So I made my tests and comparisons that way, using the old Alesis ADAT 20-bit recorder.

The above experience is as close as I can come to answering your question.

In your case, the advantage of the Accuphase T-1000's direct DSP MPX processing and AES/EBU (or SPDIF?) output might be worth it despite its 16-bit 44.1kHz rate? It would be an intriguing experiment I'd like to run myself.

-Greg

//*****************************************************************

Yes, Greg, what you say is direct reflection of my experiences. David Obergoenner is correct there is no justification to have more than 16 bit and 44K for recording dynamically and bandwidth restricted FM broadcasts. I intellectually agree but practically I am not sure.

A few years back when I was experimented with idea of FM recording I was playing good FM broadcasts into my amps and compared it to a chain with tuner-A/D-PC-D/A-amps. I was shooting to the situation where the sound from tuner direct and the sound from digitalizing chain would be absolutely not distinguishable, and I am taking in context of quite serious audio installation the can handle some complex aspects of audio reproductions. So, despite my using some very offensive AD and DA possessor I was not able to get good result at 16 bit. I had much better sound at 20 bit and 44K and I ended up with 24/88K. The Accuphase T-1000 is SPDIF 16/48K but on another side it has one less A/D processor and one less output stage – go figure how it might be. It would be interesting to know if they have more than 16 bits after multiplexing. If tay do then whatever the Accuphase T-1000 sound not that it might sound much better (I am sure my D/A is better then what is in the stock Accuphase). So, perhaps someone shall start to raise this issue and in a year or so Accuphase would do the Makr II with 20 bit resolution?  THAT would be very interesting… Unfortunately the US Accuphase representative is virtually dead and looks like he is a clueless Moron. As a result the user's interests are not well presented with Accuphase new tuner… Anyhow, I still contemplate if I need to drift toward the digital multiplexing and digital out. The idea is very lucrative and very elegant, I juts would like to have 20 bit for sake of intellectual comfort. I like different multiplex decoders, I have 4 very high quietly sounding decoders and I would have more if I know anything better is available.  Since I record FM to digital the idea of having digital after UHF, after detector, I think is very promising. Just give me 20 bits out….

Romy

//*****************************************************************

Hello,

it is a very interesting question.

In my opinion 16 bits should be sufficient,in fact I think this is a lower boundary for FM digitalization. I merely agree with your opinion that FM noise will works as a "natural dithering" in other words it will acts as the "noise smearing" technique "modulating" the rough nature of the 16 bits. A similar technique plus noise shapping and versampling ones was used by PHILIPS in their low cost 14 bits DACs to increase the resolution to a 16 bits equivalent one.

A very similar technique is Super Bit Mapping in CD leading to 18-20 bits equivalent resolution. If I remember it well the optimal "smearing" noise amplitude should be approx. 3 times higher than LS bit amplitude. If we get as a maximal FM tuner reachable S/N ratio of 87-90dB(A-weighted)and as a CD maximal S/N ratio of 100dB(A-weighted)we can see that the difference between them is nearly the same these 3times(10dB). That's why I think the 16 bits are sufficient, of course this 16 bits
stream must be then processed by using of adequate oversampling and noise shapping techiques to get a good result.

Best regards,
Cvetan

//*****************************************************************

I would not use any less than 24bits/88Ks/s for quality stereo recording, but you already know this

My take on this is that EVERY A/D conversion introduces a loss of audio quality for various reasons.

The first question is what is the actual effective number of bits in the A/D convertor?  I get around 18bits out of my cheap Crystal delta-sigma chip, you get close to 23bits with your best pro convertors. The T1000 is around 19 or 20bits.

The next question is what is the bit width of the data path used in the T1000? I doubt its better than 24bits. Any Accuphase engineers around to counter? You need at least a 48bit fixed point path to maintain sound quality (Rane used to have a app note about data path width on its website)

DSP is funny in that round off errors can be exacerbated by different math operations. So you need a very large data path all the way thru processing until the last step or D/A conversion. Then you choose from various dithers,filters ,etc. All of those leave a distinct sonic fingerprint when downsampling to 16bit. On top of this, all processing must be real time which limits smple rate and resolution. No thank you!

What type of input filtering is used, slow roll off or sharp? This is fixed by the AD chip

How is the AD chip clocked? Dedicated low jitter clock or slaved from DSP PLL clock (YUM!)?

Ideally you could bypass all these internal "compromises" by recording the composite signal directly from the Pulse Count Detector. This would require 1 channel of quality external 24bit/192ks/s A/D conversion. You would then "post-process" the data into stereo using the massive DSP power of modern GPUs (nvidia CUDA for example). I know of one DIYer that is using CUDA code for multichannel 4 way crossovers in REAL TIME so post processing to only 2 channel MPX would be much faster, maybe even 5-10X real time. GPUs typically have 128 bit data paths so data integrity and avoidance of overflows is much better than a 48bit DSP

If you code it, then you can choose resolution, sample rate,dither functions, and filters that work best with YOUR D/A convertor.

Of course if you have managed to get this far you don't need a T1000, could use the detector out of a TU-X1

Please keep in mind that I am limiting this "extreme overkill is no vice" type discussion to the very best programming and station quality. There is no point in doing this for commercial radio! Theoretically, 14bits/38ks/s is what generally considered "FM quality". So much for theory..if you want that then get a Sony hd radio.

-J

Rerurn to Romy the Cat's Site