Urgent: Sample Accurate PCM Not Working, NH900

[daniel_cello]

Hello friends,

I have a job coming up tomorrow where I will be recording a singer and a piano player. I will capture the sound in pcm to a compact flash recorder (m-audio microtrack) with my nh900 running as a backup, also in pcm. As a preparation I made a digital recording from a cd to both the recorders and compared the results to the ripped track. I had no problem getting sample accurate results on the microtrack but my md is giving me trouble.

I have tried what the manual suggests (setting recording volume to auto(agc) in rec-pause mode and then starting the recording) and I have tried with manual recording level of 23/30 and 24/30 which gives values lower and higher than the original. I have double checked that I am recording to pcm. What else could be affecting the bitstream on the optical input before it gets written to the disk?

There is of course the issue of not being able to save digitally made recordings to wav in SS3.2 (which I have and is downloaded off the forum.) To solve this I am running total recorder set to 100% recording level, but there still is uncertainty that I am saving a sample accurate copy of what SS is generating.

I have read this thread too, but I doesn't give sample accuracy, only approximate level matching. The samples could be off by +/- 50.

The md is just a backup solution, but I really would like to make the best effort possible. Any ideas?

Thank you in advance,

/Daniel Enochsson

[e1ghtyf1ve]

*puts thinking cap on*

Now this is a tough one. Let's get the easy stuff out of the way first. Sonic Stage 3.3 is out and I believe it is the least buggy of the bunch, even better than 3.2. So I would start with that as a baseline. Next, for converting recordings made from optical-in to WAV, you will need the excellent HIMDRenderer program from

MarCNet 2

I'm not sure the program is bit-accurate, you'll need to check with the author on this. But it should be worlds better than using the Total Recorder kludge.

What kind of hardware do you use to feed the MT (coax) vs. the Hi-MD recorder (optical)? Lunatec? Help, I'm stumped...

Cheers

[Roamer]

I remember from another thread Marc telling than his software is almost, but not completly bit-accurate.

[daniel_cello]

Thank you for your suggestions!

I will try the HiMDRenderer and see what gives. If the "errors" occur when doing the recording or when SS is importing the song it won't help, but at least there is one less thing to worry about.

To split the signal I use a small box called CO2 made by Midiman. It has three modes:

• take an input signal on optical and output it to both coax and optical
• take an input signal on coax and output it to both coax and optical
• convert an input signal on optical to output on coax and vice versa (bidirectional)

Cheers,

/Daniel

[ozpeter]

When recording digitally via the optical cable, level controls do not function, so I am mystified by your report that you've made changes in the recording by changing the manual recording level.

[greenmachine]

They do.

[ozpeter]

Heh... well they shouldn't....

Here, comparing a CD rip of a track with an NH900 optical in recording of a track (with no regard to control settings). the two go out of sync quite quickly, which of course they shouldn't. So the transfer from CD to NH900 PCM is not bit accurate.

I compared rips of the CD using two programs, and they are identical, which indicates it's not the CD ripping at fault.

I compared a Total Recorder transfer with the latest version of the HiMDRenderer program's conversion to wave, and apart from a single sample part way through the 3 minute test file, they are identical.

This points pretty conclusively to the NH900 being at fault.

I'll play with some settings to see what difference it makes, but I guess Sony didn't anticipate anyone getting from their optical in > wave, so didn't bother with bit accuracy.

Hmmm, shows I should read the manual. Indeed optical in recordings are affected by level etc settings, which I never realised. I've recently made a number of optical in recordings as backup to other media and never had cause to check the results, which are likely therefore to be somewhat iffy...

Under the circumstances therefore I wouldn't expect a bit accurate transfer via optical in.

[dex Otaku]

1 - As stated by Sony's engineering division [it's in one of the docs in the research section on minidisc.org, unfortunately I can't recall which as I read it probably a year ago], all MD and HiMD recorders resample their input [not just reclocking, but resampling], even from the optical input, even when the sampling rate is already 44.1kHz. This makes absolute sample-accurate copies completely impossible, but doesn't necessarily mean a trashing of the quality of what's coming in. Any allusions to "fault" are themselves faulty, since this is done on purpose.

2 - The microtrack can record in 24/96 - I assume if you're trying to make sample-accurate dupes directly from it, you're recording on it in 16/44.1, but you might want to check this to make sure [the HiMD itself shold not be able to recognise anything at a rate higher than 48kHz in any case, so it's doubtful that this is the problem]. If you really want to check the by-word accuracy of the microtrack, plug it into something that you know does no reclocking or resampling, and check to see exactly what it's putting out [i.e. 16/44.1 if that's what you've set]. If you're recording in higher than 16-bit res on the microtrack, the HiMD recorder is likely truncating the incoming words to 16-bit if it recognises longer wordlengths at all.

3 - A difference of +/-50 [out of 65,535 plus sign bit] is awfully small, if that's what you were measuring by at least.

4 - While I'm pretty sure that manual levels does work over the digital-in on HiMD [i have never had the oppotunity to try it, myself], if you really want accuracy, don't use it. Having it disabled [the default] should copy exactly what's coming in, level-wise, keeping in mind that everything is resampled.

5 - Totally a side-note - I have run both my NH700 and RH10 to make a 4-track recording [from analogue sources] to be manually sync'd later, and noted that their clocks did appear to wander relative to one another, even taking into account the fact that Sony's implementation of auto-trackmarking while using line-in [which can not be disabled] causes tiny repeated sections of inconsistent length where track marks occur. The end result was that my two [PCM] recordings had to be manually resync'd multiple times over the duration of the whole recording to reflect the drift.

6 - the microtrack might also be watermarking its output, though it's highly doubtful; there has been speculation that HiMD recorders might be watermarking their input as well, which, given Sony's track record, is slightly less doubtful.

Cheers.

[ozpeter]

I'm not having a good night.... first not reading the manual, then not reading the older very handy thread that daniel_cello mentioned. What dex now says ties up with my tests, then - there will be drift time-wise, but if you just bang the MD into record the default AGC ON setting will actually provide equal (well, near as dammit) levels. Unless needing to actually achieve perfect sync timewise with something else, for all practical purposes the recording should be fine.

My precious recent backups having thus been recorded, they should be ok. Whew.

My backups are to Hi-SP anyway..... so there goes bit-accuracy anyway.

[e1ghtyf1ve]

I'm not having a good night.... first not reading the manual, then not reading the older very handy thread that daniel_cello mentioned. What dex now says ties up with my tests, then - there will be drift time-wise, but if you just bang the MD into record the default AGC ON setting will actually provide equal (well, near as dammit) levels. Unless needing to actually achieve perfect sync timewise with something else, for all practical purposes the recording should be fine.

My precious recent backups having thus been recorded, they should be ok. Whew.

My backups are to Hi-SP anyway..... so there goes bit-accuracy anyway.

I wouldn't worry about it too much - this "taper mentality" will remain a mystery to me, at least. Digital sampling in general (yes, even 24-bit) always throws away data coming in from analog sources such as live performances. The key here is can the human ear (your ear in particular) detect the difference? Watermarking should only be worrisome to me if I were doing something illegal - it's as meaningless as debating how many angels can dance on the head of a pin... But then I'm a mental midget. To me, the mics used make all the difference

Cheers

[dex Otaku]

Digital sampling in general (yes, even 24-bit) always throws away data coming in from analog sources such as live performances.

Um. Wha? Can you elaborate on this a little, please?

Digital sampling is not a form of data-reduction. It is sampling. Past a possible low-pass filter used to elimnate distortion by frequencies above the Nyquist freq. of a given sampling rate, whatever signal comes in to an analogue to digital converter should be recorded [i.e. sampled] to the best of the ability of that ADC. No data gets "thrown away" in the process, though some is purposely filtered out, and the process itself always has limitations that cause [mostly] well-known types of distortion.

Data reduction, i.e. lossy compression, is another matter entirely, but we're talking about straight, uncompressed, lossless PCM recording here, not recording with lossy compression. The resampling issue and its side-effects aside [which should exhibit about as much change in the recorded signal as adding dither does], what goes in should also come out.

I'm interested to know where you came by this viewpoint, or if I'm mistaking the context you've said the above in.

Cheers.

[e1ghtyf1ve]

Um. Wha? Can you elaborate on this a little, please?

Digital sampling is not a form of data-reduction. It is sampling. Past a possible low-pass filter used to elimnate distortion by frequencies above the Nyquist freq. of a given sampling rate, whatever signal comes in to an analogue to digital converter should be recorded [i.e. sampled] to the best of the ability of that ADC. No data gets "thrown away" in the process, though some is purposely filtered out, and the process itself always has limitations that cause [mostly] well-known types of distortion.

Data reduction, i.e. lossy compression, is another matter entirely, but we're talking about straight, uncompressed, lossless PCM recording here, not recording with lossy compression. The resampling issue and its side-effects aside [which should exhibit about as much change in the recorded signal as adding dither does], what goes in should also come out.

I'm interested to know where you came by this viewpoint, or if I'm mistaking the context you've said the above in.

Cheers.

This is getting off topic, but I'm so glad you asked! Aside from the definition of data-reduction, I agree with everything you say. Let me explain.

I'm well aware of the distinct difference between lossy software/hardware implemented data compression schemes and digitally sampling an analog signal stream. However, both (in the audio domain) make use of limitations in human perception. "Throwing data out" or "filtering" is, in my opinion, really just two sides of the same coin. The bottom line is that no matter how high you make your sample rate and your bit-resolution, all you are doing is changing the characteristics of your low pass (or actually band pass) filter. Once you are in the digital domain, I agree of course, if you don't have to resample, what goes in must exactly come out. But in my book, this is a serial bit-stream copy process, almost like copying files on a hard drive, and has nothing specifically to do with an actual format such as PCM.

So is the "watermarking" audible? Not to me, it isn't.

I'm unequivocally against DRM. But if I can't hear any audible degradation, even more, if I subjectively determine that the DRM crippled device actually sounds better than it's unfettered siblings (like Hi-MD vs MT), I'm not going to let DRM stop me, especially if there exist splendid workarounds such as HiMDRenderer.

Hopefully nobody got offended by my opinions - I can't help but chuckle when I read about and hear the results of some expensive equipment recording the distortion blasted out by PA systems...and the discussion about how 24-bits sound better. It's just priceless, this taper scene mentality will not fit in my little brain, that's for sure.

Cheers

[dex Otaku]

Okay.. thanks for elaborating. Agreed, though again - the watermarking is pure speculation.

As for 24- vs. 16 bits, to me it's not so much that high-res souns better, it's that it lasts longer through editing cycles with less degredation through successive passes of processing. Properly-exploited, 16-bit audio is fine to me; I'd take as much resolution as possible in the original conversion, though, just for the sake of reducing error as much as possible, at least until the final stage before distro.

I tend to edit even 16-bit audio in 24-bit mode, with DSP as high as the res of each plugin allows, with a final pass of requantising and dither before it hits an audio CD. I encode most lossy distro sources from dithered 24-bit data. It may seem like overkill, but I've made it a habit, and there are times when the difference is noticeable.

Yeah, we're wandering off-topic, but thanks for the interesting sub-discussion nonetheless.

Cheers.

[e1ghtyf1ve]

We're on the same page here, terrific. I just wanted to clarify my stand on 24-bit resolution - of course it often leads to better results, but I should have been clearer and written that it is all too often wasted on recording the distorted output of PA systems. Oops...Cheers

Edited November 14, 2005 by e1ghtyf1ve

[ozpeter]

My concern about my backups was simply that the levels may have been totally wrong - though even with the tiny meters on Hi-MD hardware I'd like to think I would have noticed at the time.

To me, the mics used make all the difference

Indeed. I use Sennheiser MKH series mics as the starting point, and that kinda helps.

Edited November 14, 2005 by ozpeter