This one does not really seem to have a home on the forums, but Phono seems quite an active place with guys who know their stuff. BTW I'm new to this forum and I'm based in the UK.
I'm into trying to record my old LP's to digital - and trying to stick to HiFi.
Had a bit of a struggle too- old system, bought c. 1978, is:
PL12D tarted up - oiled, new belt, rejig wiring, go-faster stripe Shure M75ED - new stylus, re-aligned etc Yamaha Natural Sound CA-610
Had earth loop problems (hum + sub and over-harmonics of 50Hz up to 400Hz) so got out big hammer to fix.
All above items now run at 110v via a step-down isolating toroid - everything floats and mains carried noise, surges etc also get 1/2 too on step-down.
Zero hum now...
The Yamaha's Record Out runs into a PC computer A/D system - and that's my beef for this post.
(I will not go into 9Hz rumble on PL12D just yet)
Now as far as I read around the web, everyone seems happy with sampling at 41 or 44K samples per sec (sps) - but to me that seems crazy. It just can't give HiFi! To my mind; then I could be wrong. Am I?
Here is my rationalle:
Nyquist (spelling?) has a magic formula which says sample rate must be at least x2 top frequency.
OR restated, if sampling rate is less then x2 top frequency... you are guaranteed to miss the signal.
But at over x2 you can still miss it; it just that you now have a chance of getting a something, but no actual guarentee you will get anything.
This is simply shown.
Take a sine wave and sample it at the peak + and peak - values. This is what is in the diagrams showing how sampling works.
The sine has done a cycle and you have sampled it twice.
What are the chances that you started sampling at the right point? Why at the top of the signal?
It's random where you sample vs. where the singal has got to. It is quite possible the samples get taken at 180 and 360 degrees; that's allowed and just as likely.
But samples at 180 and 360 give you a straight line! Zero signal (or 100% distortion in THD terms)!
Increase the sample speed a bit and you get captures which beat from full strength to zero and back to full signal again, the beat frequency being the difference in sample speed and x2 the signal.
I've reproduced this; the effect is real and it sounds awful!
Hm. As I read it, Nyquist and sampling was about information theory, not audio fidelity. OK so what to do? Perhaps ... figure out a better sample speed.
To my mind, the way to approach this is to try to capture the signal with fidelity i.e. no errors.
And the minimum samples I'd want are, expressed around a sine wave:
0 degrees -the starting point. 45 degrees -the "halfway-up a slope" value 90 degrees -the peak positive signal value 135 degrees -the "halfway-down" value
The sequence repeats "upside down" for the signal's negative section; the 180 degree value is the start point for the 4 samples on the negative side.
So. 4 samples in each half-cycle; that makes 8 per cycle.
Even then THD I expect to be high (but really really difficult to work out). The digital to analogue converter will invent straight lines or some roll-off curve when playing back between sample points, missing the actual waveform that should be there.
Oh dear - this means to record an LP in some "BetterFi" we need to capture say 18KHz at 8 samples per second; that's ahem 144,000 sps.
Not 44,100.
Conversely, what's 44K samples per sec good for? 5.5KHz ..oops.
I have been playing with an external sampling ADC unit (it's external to the PC case) at both 48K and 96K samples per sec. And when looking at the waveforms on-screen there IS detail in the 96K waveform that the 48K version misses out.
Even so, at 96K the suite can only capture up to 12KHz ish with some sort of fidelity.
And the recordings sound... not so much as brighter as more ..er liquid, detailed.
So it seems to me.
Is this analysis right?? Are my ears deceived?
It seems I'm flying in the face of accepted wisdom with all this; but then I do that sorta stuff.
steve PS where should a post on recording/digitising go?
Hard to tell since I find it hard to decipher what your actual question is. If you are asking should you sample at the highest possible frequency; the answer is yes. If you are asking will the fidelity be better at the higher sampling rate; the answer is yes. If you are asking why you still hear distortion at 96kHz sampling; I don't know the answer to that question.
I'm not sure how you determined that 96kHz can only capture up to about 12kHz in response "with some sort of fidelity". Or, how you determined the rumble to be at 9Hz.
My answer, at this point, would be to check with the software manufacturer for your PC's A/D system. Sorry I can't be more helpful.
I think I am with you, Steve. Though, to be sure, I'd need a diagram or two.
What has always puzzled me about Nyquist theorem, a term which crops up on this forum from time to time, is how the wave can be completely reconstructed from samples of amplitude that are taken at a frequency only twice that of the frequency of the wave. This is what people claim.
If I understand the first part, you are sort of asking how do we know the randomly-chosen time of the first sample does not correspond to, let us say, 0º. Then, the next sample, at twice the wave frequency, will be at 180º. For a sine wave, the amplitude at both of these times is zero. So we have sampled at twice the sound frequency, as Nyquist says; we then reconstruct the wave from the samples, and we get silence.
That's my conceptual problem number one, and I am looking at this in a very qualitative sort of way, and could have missed the point.
If I understand, you are going further and pointing out that you get different reconstructed waves according to where the sampling starts. It seems to me one will only get the original sine wave (a) if you know it was a sine wave and not something else and (b) if your first sample happened to be at 90º. I suppose one could do multiple sample trains starting at random times and integrate those, but one then has sampling at more than 2 x the wave frequency. Intuitively I would expect the accuracy of the wave to go up with something like the square root of the number of samples - I'd have to sit down and think beyond my current level of maths to be sure.
In contrast, you are suggesting sampling not at random, but at pre-determined points in the cycle. Yes, I think eight should do it.
How am I doing?
Then I have another problem - how do we know it is a simple sign wave? Real musical sounds are much more complex. Anyone could easily draw a wave that is the sum of a number of harmonics with or without the fundamental, giving the same values as a simple sigjn wave at each of your eight samples, and yet not be a simple sign wave at all. How does the digital sample distinguish the two waves? Nyquist, even with 8x not 2x, only works for the fundamental frequency, it seems to me.
In practical terms, to satisfy your 8x, not 2x, sampling, we would need sampling 8x the highest sound frequency. To allow for those who can hear to 20 kHz, we need 160 kHz sampling. Leave some space for a filter, and we might go to 8 x 22 kHz equals 196 kHz. This is four times that of CD, and a bit beyond the sampling frequency of DVD-Audio (192 kHz).
So, we end up with the conclusion that CD was a flawed format, all along, not as good as analogue (choruses of disapproval from many directions) and not at all "Perfect sound" as those guys originally swore it was.
Have you tried DVD-A or LPCM at 192 kHz?
I understand one can get ADCs for audio recording that will do that, computers have no problem with that these days, and one can write DVDs.
There was an article in HiFi News on how to back-up one's LP collection in digital format and I think it came to the same conclusion. It was some time late in 2003 or early in 2004, I think.
There is also the DSD/SACD school of thought, of course. The practical problem, there, is that the licence holders will never allow a domestic writer. Their mission is primarily maintaining ownership of content, it seems to me, and I do suspect SACD was designed for encryption, and with that objective firmly in mind.
I have no idea where this thread should be filed, but "phono" seems as good as anything. All those years of people replacing their LP collections with CDs!
Thank you for your replies and interest in my rambling post.
Jan, I don't think I can nail a specific sample period as "right" for a given signal. But I'd like to think that there were enough samples to show the right waveform - still a moveable feast; it'll vary from waveform to waveform with no right answer.
As to how I knew I had 9Hz rumble from the PL12D - simple! In Audacity, I counted the LF cycles in a "quiet" section of record and worked out the frequency. Yep could be something else then rumble - but what??
Always problems.
Right now, reviewing recordings, the "s"'s are to drawn-out and false-silibent. Not crisp where the heck is this going wrong?
Tracking at 1 gm with 1 gm anti-skate; stylus set up to better-than 1/2 mm of where it should be according to built-in overhang checker.
A point to note. Do have a 1980's HiFi setup disk; this has samples of all sorts of setup tones and test freq. and it tells you what you should get for each track.
So, looking at the waveforms for these tones etc, I can SEE bias / skating errors! And see the deck runs at - 0.14% speed; a tad slow.
Anyhows it looks as good as I can expect; R channel clips at peak + sooner then L. Make sense?
It's good to have a viewable tool, showing you what's going on with the output.
I'll play with this some more, increase desk then platter local dampening some (add weight) and keep an eye out for rumble. Fortunately Audacity allows nice notch filters.
where the heck is this going wrong?