Measuring phono stage RIAA accuracy with a computer

[gpapag]

Hi George,
How did you produce your smooth Riaa corrected curve from Scott's IRiaa curve.

Hi Hans
OK we both cheated a bit (no hardware use, only software) but it’s only for demonstration.
You LTSpice, me Audacity

But I have a problem in understanding RMAA, because when taking a 32K FFT from Scott's file,

It’s not an issue with RMAA.
As Scott has explained in his LA Vol. 10 article, these 3 sec long files are optimised for 64k & 96k FFT and no window function.
If you run them with some window function other than the rectangular, the FFT envelope looks equally noisy regardless of FFT length used.

George
>Edit I see Scott answered this

 

[jaddie]

Well no actually. One doesn't doesn't have to pull too hard at the abundant loose ends in archive material to realise there's still plenty that lacks proper explanation, and plenty of uncharted water.

You obviously don't know me or the body of work and time I invested over the years in understanding vinyl playback technology.

Well, think again. There are plenty of 'known' anomalies in classic f response which still lack proper explanation, and vary with level.

Disagreeing is your prerogative. But the mechanisms involved are self-evident and obvious. You could, of course spend the best part of a decade researching, thinking through, testing, posting thousands of results, and measuring as I have. Take a look.

Anyways, I don't mean to derail the thread, and I'll bow out now. I made my point that it's pointless to go overboard on designing or measuring very accurate RIAA preamps. But hey I don't deny it's fun, and I've been there and done it.

LD

All of that, but you won't share it?

 

[Hans Polak]

Hans the maths don't work that way, the files are set to work at a single sampling frequency without windowing. Decimation by two (using 32k FFT, has mathematical artifacts from imaging/aliasing). I keep having this problem with making files for folks.

Hi Scott,

IMHO this is not how RMAA is supposed to work.
The program takes a time window of X samples all at the original sampling rate of the file offered, where X can be anything between 2^7 and 2^18 and produces a X bin FFT.
So there is no decimation causing aliasing involved.
After the first FFT has been calculated, the time window shifts 50, 25, 12.5 or 6.25% of X and takes the next X samples to calculate a new X bin FFT and so on and so on until all time samples have been used.
Then follows averaging off all calculated FFT's, resulting again in a smoothened X bin FFT.

So when offering a 96Khz file, no matter what X is, at the end the FFT always shows a bandwidth of 48Khz, proving that no decimation and aliasing has taken place.
With X = 65,536, resolution is 1,5 Hz and with X=4096 resolution is 23 Hz. That's the difference it makes when choosing X, but the FFT bandwidth stays always 48 Khz in case of a 96 Khz file.

So when going from 65K to 32K, there is no logic why suddenly the spectrum is less smooth where the filter bandwidth has doubled from 1,5 Hz to 3 Hz.

In the Image below I have processed your IRiaa 96Khz file with different values for X, all FFT's calculated without a shaping window and with an overlap of 50%.
As you can see all FFT's are showing a bandwidth of 48 Khz, so no decimation has taken place.
from X is 32K to 4K you see the noise in the image diminishing as expected, because the filter bin becomes wider, but for X is 65K that should show the highest noise, spectrum is suddenly noiseless.

So I can only repeat that I think that in this case the 65K FFT is not reliable and has most likely to do with a bug in RMAA, unless you have a better explanation.

Hans

 

[scott wurcer]

So I can only repeat that I think that in this case the 65K FFT is not reliable and has most likely to do with a bug in RMAA, unless you have a better explanation.

Hans

No bugs just the math, all your artifacts can be predicted. I keep running into this, this file only works at 64K unwindowed FFT's period what's the problem it does not work at all at any other length FFT?

 

[billshurv]

Here's my problem with this. Every other vinyl-related anomaly has been documented to death, tracking angles, VTA, stylus shape, and dozens of others. Yet, you say this one flew under the radar? Then how do you claim to know about it? If it caused an error of several dB, to the point where it was clearly audible, I think it would certainly be documented.

Now I am about 8 years behind LD in looking at these issues (and at least 50 IQ points), but it does appear that a whole load of stuff was not published even if companies new about it. I discovered only last week from a story being told by someone berating me for having the audacity to question cartridge resonance models that ADC had worked out that cantilever flex was real around 1980. They never published because they figured it gave them an edge in the market. Sadly journalists all went MC, so the ADC brand is almost forgotten.

Looking at magnetic non-linearities this was discovered and various solutions implemented, but the fact that it might be level dependent was glossed over. Shure had their laminated pole pieces in the VSTV, AT have their paratoroidal generator and ortofon have their split pole piece. Generally they all marketed these as 'eddy current reduction' and left it at that, although the ortofon 500 series brochure did have some interesting graphs.

So this means that you should be able to compare and ortofon OM (solid pole piece) generator with a superOM (split ring) and measure something useful. If you had the right test tracks at least....

 

[kevinkr]

This assertion took me quite by surprise when I read it earlier today, I would think this effect would be quite subtle compared to gross response errors introduced by undesired compliances, etc in the mechanical system which no one led us to believe were level dependent.

This would certainly argue for a few different levels of pink noise on the test disk which would allow us to confirm or disprove this as a serious abberation. (I don't doubt there is some level dependent variation in response but I would hope within the dynamic range of most vinyl this is not a big factor)

LD stop giving me reasons to badmouth vinyl! Bill has already heard quite a mouthful from me about other vinyl pb problems..

 

[billshurv]

Kevin: you are plying such a different groove I always love to hear about your trials and tribulations

But seriously I think this thread is covering some very useful ground. Covered before but work repeating as it can be daunting first time to work out how to measure this at home. Thanks to the work of people on diyaudio the hard stuff is done and its just implementing it.

 

[kevinkr]

For those who don't know me as well as Bill does, the different groove he refers to is currently a pile of Panasonic strain gauges and the funky electronics (mine) that makes them work, in addition to several air bearing linear tracking tonearms from ET.

Basically I am orbiting another planet and there is no going back. Getting here has been a bumpy ride to say the least, and there are more trials yet to come..

 

[jaddie]

Now I am about 8 years behind LD in looking at these issues (and at least 50 IQ points), but it does appear that a whole load of stuff was not published even if companies new about it. I discovered only last week from a story being told by someone berating me for having the audacity to question cartridge resonance models that ADC had worked out that cantilever flex was real around 1980. They never published because they figured it gave them an edge in the market. Sadly journalists all went MC, so the ADC brand is almost forgotten.

Looking at magnetic non-linearities this was discovered and various solutions implemented, but the fact that it might be level dependent was glossed over. Shure had their laminated pole pieces in the VSTV, AT have their paratoroidal generator and ortofon have their split pole piece. Generally they all marketed these as 'eddy current reduction' and left it at that, although the ortofon 500 series brochure did have some interesting graphs.

So this means that you should be able to compare and ortofon OM (solid pole piece) generator with a superOM (split ring) and measure something useful. If you had the right test tracks at least....

You're zooming into the specific, I'm still at the long shot of general. IF there is a "significant" change in FR with level, that's something that absolutely could be known and documented. What you and @luckythedog are saying is mildly interesting, and maximally frustrating because you imply a "significant" FR issue, but don't even bother to define "significant", not even +/- 1dB.

So, the first question is, "how big a problem is it?"

Second question is, "how was the problem discovered and measured?"

The third question is, "Is it consistent across all cartridges?" (I'm going to assume "no", but let the "experts" speak).

The fourth, possibly biggest, problem here is, LOTS of claims, NO data, NO references, and so NO substantiation. There's the conspiracy theory angle too. So why should anyone believe any of this...exactly?

You guys present an problem, but how is anyone supposed to respond to this? It sounds like audiophile nonsense, and has all the ear marks. My nearly half-century experience has managed to somehow steer clear of the entire thing, though I'm not the most proficient expert in the field. I've measured lots of carts, lots of phono preamps, played with loading, physical cal, etc., and found lots of issues along the way. But I've also done a project where I had lacquer masters cut "barefoot", without any special mastering, EQ, processing for vinyl, etc. Two of them. They came from digital masters. I could then take a precisely calibrated cart, arm, and preamp, match levels and do an actual ABX comparison to the digital master the lacquer was cut from. What I found was, surprisingly little spectral differences, differences in noise, and a bit of separation, but otherwise a surprisingly dead-on match. The material was fairly dynamic too, so if this level-induced FR change was "significant" in my world, I'd have heard it. I did not.

So, what the heck are you guys on about? Post the data, your tests, anything but wild unsubstantiated claims backed only by "you don't know me, but I'm really smart!" kind of nonsense. Perhaps you are smart. Perhaps you do have data. Perhaps this is real. And I don't know you! But how am I supposed to put faith in someones random statement on a public forum in light of all of this?

And...just to scale this...analog tape does have this issue, different mechanism, but on the order of possibly 1dB - 3dB, machine, speed, and tape formulation dependent. Successive generations make it worse. And that's what everyone mastered with back in the day. Then we have cart loading effects, and general RIAA miss-tracking. And on and on. Yeah, it's a flawed system.

But if you have some hush-hush, secret society-level 3dB response excursion to let the world in on, then how about it? I, for one, welcome the opportunity to learn, but I'm not going to drill through the earth's crust to find the answer.

I've now requested detailed info 3 times, nothing received so far. Not even "it's about 1dB at 20kHz with a stiff west breeze on the second Tuesday of next week" kind of data. Nothing. This is NOT helping your position!

 

[Conrad Hoffman]

IMO, if there's a change in frequency response with level you'd probably see it as a bad THD measurement at the frequencies where the change was occurring, say at 20 or 20,000 Hz, since it's a non-linearity.

 

[Hans Polak]

No bugs just the math, all your artifacts can be predicted. I keep running into this, this file only works at 64K unwindowed FFT's period what's the problem it does not work at all at any other length FFT?

Scott,
It's no real problem that your IRiaa file does not work with other length FFT's, but I simply don't like to use tools things that I do not understand and none of your earlier indications of decimation, aliasing and windowing did apply.

White noise is still "Noisy" when taking an FFT of whatever length.
So the fact that the FR at 65K of your file is noiseless, probably means that most likely to construct the .wav files you have followed the way back with an inverse 65K FFT from a flat noiseless frequency curve.
In that case the slightest alteration like windowing or using different length FFT's will inevitably result in the symptoms I have seen in RMAA.

I'll read your article in LA10 that George mentioned and see if this brings me a step further.
But to my great joy I saw in said article that I was as #7 in your reference list

Hans

 

[Hans Polak]

IMO, if there's a change in frequency response with level you'd probably see it as a bad THD measurement at the frequencies where the change was occurring, say at 20 or 20,000 Hz, since it's a non-linearity.

What about at least 1% distortion at fair levels.
That could easily be caused by some (mild) level of level compression.
LD never said it was a huge problem, but I'm also anxious to measure this and give it a dimension however small it may be.
Denying the rigoureus way Jaddie does that some issue might exist does is not a very scientific approach and does not bring us one step further.
This forum is about bringing everything to the surface and if well proven that it is insignificant, it is early enough to make the next move and else we have enrichened our toolbox.

Hans

 

[ctrlx]

providing you can get suitable software for Linux, setup/calibrate it, then its relatively easy to measure the response of an electrical circuit.


attached is the response of an riaa cct after I added warp filtering, taken with speakerworkshop. (yellow trace is the reference)

 

[billshurv]

@Jaddie. I note from your post history that you requested an 02 board 7 years ago then came back recently with a fairly agressive attitude. Chill out, we are all friends here.

In terms of the magnitudes we could be talking about take a look at the attached which is from the ortofon marketing bumph for the 500 series which were the first split pin generators. X-referencing with other marketing stuff suggests 1dB per division so the solid pin, with optimal loading gives a 3dB dip in the HF region, which they put down to eddy currents. Very few attempts have been made to find an electrical model that matches that response to then look into deeper issues such as level variance. Combined with the fact that test records don't generally have calibrated variable level sweeps these days and you can see why this slipped until the radar. Also doesn't help that journalists don't seem to believe that MM can give the same performance as MC.

For me it's interesting things to investigate even if the end result is a list of the most technically excellent MMs out there and how to interface with them without actually solving all the issues. All good harmless fun.

 

[mkc]

taken with speakerworkshop. (yellow trace is the reference)

Speakerworkshop is windows as far as I remember?

Mogens

 

[gpapag]

Erratum:

But to my great joy I saw in said article that I was as #7 in your reference list

Should be:

But to my great joy I saw in said article that I was as #7 in your “Further reading” list

I simply don't like to use tools / things that I do not understand

I admit that except of a very few tools (spoon, fork, knife), everything else I use is kind of alien technology to me.

George

 

[scottjoplin]

Where would we be today if people only used tools they understood? Still sitting round the campfire? Although that could be called a tool too

 

[scott wurcer]

But to my great joy I saw in said article that I was as #7 in your reference list

Hans

You are correct the pseudo random noise .wav's are generated in a standard way from inverse transforms of random phase equal amplitude data. Sorry if there was a misunderstanding but the issues with changing the sample rate do end up being from spectral leakage (windowing) or under/re-sampling. There also can be frustrating issues in a loop where D/A and A/D clocks are different.

I'm not sure I mentioned it specifically but another benefit of the unwindowed processing is that there is no spectral leakage and synchronous averaging can get extremely low noise floors. I did reference a recent paper on some new windows that were aimed at 24bit noise floors but they are not available in any tools I know of.

 

[Hans Polak]

Where would we be today if people only used tools they understood? Still sitting round the campfire? Although that could be called a tool too

Ha, ha, exactly because of that attitude we are no longer sitting around a campfire still believing in dozens of God's for fire, rain, wind, thunderstorm etc, just because we went into the why and found the reasons that made us understand the causes.
That inquisitiveness brought us where we are today enjoying our Audio Hobby amongst other things, not bad at all.

Hans

 

[jaddie]

@Jaddie. I note from your post history that you requested an 02 board 7 years ago then came back recently with a fairly agressive attitude. Chill out, we are all friends here.

Sorry. Unsubstantiated myth is my hot button.

In terms of the magnitudes we could be talking about take a look at the attached which is from the ortofon marketing bumph for the 500 series which were the first split pin generators. X-referencing with other marketing stuff suggests 1dB per division so the solid pin, with optimal loading gives a 3dB dip in the HF region, which they put down to eddy currents. Very few attempts have been made to find an electrical model that matches that response to then look into deeper issues such as level variance. Combined with the fact that test records don't generally have calibrated variable level sweeps these days and you can see why this slipped until the radar. Also doesn't help that journalists don't seem to believe that MM can give the same performance as MC.

There is not enough information about that graph to conclude anything. Assumption about vertical scale...well, it's probably ok but who knows? Test conditions are undefined, and it seems to indicate a significant change in the cart for each of the traces. Not enough info.

For me it's interesting things to investigate even if the end result is a list of the most technically excellent MMs out there and how to interface with them without actually solving all the issues. All good harmless fun.

I'm not on board with the "harmless" part if mythology is perpetuated. If it's fact, I'll be 100% behind it.