My Linear Audio article had all the historical EQ's in an appendix. The Nyquist thing is a one liner, biquad s (the selection) and the 6 coefficients. Audacity FIR's are linear phase only I don't have the energy to write plug-ins for other SDK's but offer the numerical answers to anyone who wants to.
Last edited:
Thanks Scott. I think I found what I need here: https://linearaudio.net/sites/linearaudio.net/files/v10 sw app1 table a-1.xlsx and here: https://linearaudio.net/sites/linearaudio.net/files/v10 sw Appendix 1b web.docx
Prior to finding the LA links I visited Wayne Stegall's site Website of Wayne Stegall - Digital Phono Equalization and used riaaiir to produce a Nyquist file. I modified his original 96 kHz example to add 20 dB gain using riaaiir.exe I ticked the legacy version 3 syntax box in Audacity in the Nyquist prompt box. Not sure how accurate this is but it sounds OK:
The .ny file is here: http://proaudiodesignforum.com/code/RIAA_Stegall_Test_96K_20dB_Gain.ny
Prior to finding the LA links I visited Wayne Stegall's site Website of Wayne Stegall - Digital Phono Equalization and used riaaiir to produce a Nyquist file. I modified his original 96 kHz example to add 20 dB gain using riaaiir.exe I ticked the legacy version 3 syntax box in Audacity in the Nyquist prompt box. Not sure how accurate this is but it sounds OK:
The .ny file is here: http://proaudiodesignforum.com/code/RIAA_Stegall_Test_96K_20dB_Gain.ny
Last edited:
I just ran some Audio Diffmaker comparisons between analog RIAA EQ, two different IIR Biquad plugins and Cool Edit's FIR.
I used the first 60 seconds of Supertramp's "Logical Song."
One pass was recorded RAW (flat) at 96 kHz a second pass recorded with analog RIAA EQ also at 96 kHz SR.
I then applied RIAA EQ in DSP using Nyquist and .vst in Audacity and used Cool Edit for an FIR example. All the files (32B float) were then normalized to -2 dB. Listening to the files I noticed the FIR EQ sounded different from the rest. (No surprise.)
I then had to convert all of them to 16B files for Diffmaker.
I used the analog EQ'd version as the reference and had Audio Diffmaker compare the Stegall Biquad in Nyquist, the Matthes RIAA vst plugin and finally Cool Edit's FIR.
Analog to either IIR plugin yielded a correlation depth of about 14 dB. (Diffmaker complained about a 0.25 ppm SR error so the depth likely could be higher.) The difference file was some timing error and high top end but nothing odd-sounding.
Analog RIAA EQ compared to Cool Edit's FIR RIAA provided only about a 5 dB correlation depth. As expected the difference file sounded weird.
I then decided to compare the two IIR filters to each other. The null depth was over 60 dB. (And since they were the same pass the SR error was virtually 0.)
My conclusion is that both Wayne Stegall's Nyquist coefficients and Olaf Matthes' vst plugins, both IIR, produce virtually identical results at 96 kHz that are very close to analog RIAA EQ. I can't complain about the sound of any of them.
Cool Edit's FIR just sounds odd in comparison.
YMMV.
I used the first 60 seconds of Supertramp's "Logical Song."
One pass was recorded RAW (flat) at 96 kHz a second pass recorded with analog RIAA EQ also at 96 kHz SR.
I then applied RIAA EQ in DSP using Nyquist and .vst in Audacity and used Cool Edit for an FIR example. All the files (32B float) were then normalized to -2 dB. Listening to the files I noticed the FIR EQ sounded different from the rest. (No surprise.)
I then had to convert all of them to 16B files for Diffmaker.
I used the analog EQ'd version as the reference and had Audio Diffmaker compare the Stegall Biquad in Nyquist, the Matthes RIAA vst plugin and finally Cool Edit's FIR.
Analog to either IIR plugin yielded a correlation depth of about 14 dB. (Diffmaker complained about a 0.25 ppm SR error so the depth likely could be higher.) The difference file was some timing error and high top end but nothing odd-sounding.
Analog RIAA EQ compared to Cool Edit's FIR RIAA provided only about a 5 dB correlation depth. As expected the difference file sounded weird.
I then decided to compare the two IIR filters to each other. The null depth was over 60 dB. (And since they were the same pass the SR error was virtually 0.)
My conclusion is that both Wayne Stegall's Nyquist coefficients and Olaf Matthes' vst plugins, both IIR, produce virtually identical results at 96 kHz that are very close to analog RIAA EQ. I can't complain about the sound of any of them.
Cool Edit's FIR just sounds odd in comparison.
YMMV.
You can also generate the impulse response of any of them in Cool Edit and compare to an exact computed impulse response. I doubt you could ever build an analog RIAA to get a very deep null, the problem using diffmaker is that you have no "right" answer as a reference..
The FIR technique works fine in a convolver tool like foobar's foo-convolve or BruteFIR but you need a lot more coefficients than I remember seeing in the Cool Edit example. What I saw was a very undersampled impulse response that would have to be interpolated up to the FFT size used in the filter. I forgot the exact number at 96K but it's around 56 msec of samples for a minimum phase FIR that is micro-dB's accurate.
BTW FFT convolution and literal convolution give exactly the same answer with some trivial bookkeeping to account for the circular nature of the FFT. This can be tested down to the numerical resolution of your processor. I have the file as a short .wav if you want it.
Last edited:
True.
I was surprised the null was as deep as it was (14 dB) considering that the reference file (analog) and the RAW file were recorded in two different takes. The time base error of the turntable can be heard and you can hear Diffmaker sliding them in and out of deep null. When it's properly nulled it seems deeper. The null shouldn't be varying over time. I'll compress the Diff'd file into an mp3 and post it later.
When I compared Cool Edit's FIR to the vst plugin the correlation depth was only about 4 dB. Cool Edit's EQ far less accurate than the 1% tolerance analog RIAA.
I need to find a way to record the flat RAW file and the analog EQ'd at the same time to get a fair analog vs. DSP comparison. I don't have a vst version of Audacity and need to figure out how to route it to multitrack record. I may just route one A/D into Audacity and another A/D into Cool Edit and roll both at the same time.
What I did learn is that the three RIAAs - analog, Matthes' vst and Steagell - all sound close enough to each other to be considered useable. Cool Edit's FIR is just not phase-accurate and odd-sounding.
Last edited:
If you have an MC cartridge and a flat step-up device (either a transformer or a flat pre-preamplifier), it comes natural to do this:
Use the same output channel of the step-up device to feed in parallel :
(1) the R input of your analog RIAA preamplifier and
(2) the L input of your sound card
The R Output of your analog RIAA preamplifier will be connected to the R Input of your sound card.
The stereo recording you will perform will be L channel flat (to be DSP processed), R channel analog RIAA.
Source material will be one and the same and this is very important.
Another option I have also tried with both MC and MMs is
Use a mono (lateral) record, read it with a stereo cartridge. Use the output of both coils.
One coil of the cart to a flat preamplifier, it’s output to “L In” of the sound card.
The other coil of the cart to an analog RIAA preamplifier, it’s output to “R In” of the sound card.
Even if you cater for equal loading of the two coils, this method is inferior to the first because
a. Ticks and pops are not exactly the same at the output of the two coils and you can do nothing about it.
b. There is a small time difference between the two recordings which you have to adjust. My notes say it is anything between 0.02ms and 0.09ms. Your timing pivot point will be some well defined impulse-like scratches, the smaller are more reliable to use.
Thank you for the good work you have done and made it public.
George
George,
For flat recording I agree with Wayne, I found it hard to believe how little gain worked well. What I did is make a battery powered FET follower with some old Toshiba tophat FET's <1nV at low current draw to act as an impedance transformation (thank you mlloyd!) to the mic pre-amps in a Tascam field recorder (DR60d). I could easily see splitting the path after the buffer through flat and RIAA with two recorders (which I have). The timing difference should be far better than sequential plays on TT.
I wonder if its possible to get the original (probably digital) file used for the frequency sweep on the George's HFN&RR Test Record.
Then we can use Prof. Angelo Farina's method to get the complete frequency response ... including the cutter etc ... with the best S/N ratio.
Otherwise, if we are trying to EQ the cartridge and provide RIAA, we are down to choosing between his two long but DIFFERENT
pink noise tracks on his two Test Records.Likely Cool Edit's FIR is Linear Phase like Audacity's. Truly EVIL .. even more EVIL than Scott's FIRs.
Thanks everyone for your posts.
Richard - I do want to clarify that the Wayne Stegall Audacity Nyquist biquad coefficients and Olaf Matthes' vst IIR plugins which also run in Audacity are correct. In fact, the Nyquist RIAA and vst plugins when compared to each other have a correlation null of 60 dB. The point is Audacity is capable of accurate RIAA EQ.
I ran some more tests. At the flip of a switch I can bounce the original RAW playback through the analog RIAA. This adds two conversions to the analog RIAA compare file. In this configuration the analog vs. DSP RIAA have correlation null depths around 10 dB.
I also did simultaneous recording one flat, one with analog EQ. This required two different converters (one USB) with two different footprints. Diffmaker found a 9 ppm error in SR which it had to correct. The analog vs. DSP RIAA has a null depth also around 10 dB.
In both cases the difference sound file revealed three things: (1) An unnulled "blurp" right at the beginning of the file which may affect Diffmaker's null depth measurement. (2) Slight differences in top end which is quite likely converter footprint and (3) minor differences in low end.
The minor difference in low end is not surprising: The DSP version has gain to DC, the Analog version does not have DC gain.
The biquad RIAA available in Audacity using Nyquist or vst are, as a practical matter, as close to analog RIAA as they need to be.
To get a complete picture of the amount of available null, George's mono approach is required. I can use the same converter to acquire both samples. Wire the cart mono. Flat feeds left, right is fed by analog EQ. Only the interchannel A/D differences will affect the difference file.
Other than intellectual curiosity I'm not sure its worth doing because I'm convinced that the two DSP solutions I have for Audacity are probably more accurate than the 1% tolerance analog RIAA network and the Jung/Lipschitz 1% tolerance Inverse RIAA network used to check it.
Sonically the analog and DSP RIAA (if done right) sound virtually identical to me.
EDIT: The difference file comparing the Wayne Stegall Nyquist biquad and analog RIAA (using different convertors) is here: http://www.proaudiodesignforum.com/...ence_File_DSP_Nyquist_RIAA_vs_Analog_RIAA.mp3 (320 kbps mp3 approx 1.2 Mb)
The original files were normalized to -2 dB. I edited the uncorrelated portion at the head out. The peak level of the difference file is about -18 dBFS so the two are very close.
Richard - I do want to clarify that the Wayne Stegall Audacity Nyquist biquad coefficients and Olaf Matthes' vst IIR plugins which also run in Audacity are correct. In fact, the Nyquist RIAA and vst plugins when compared to each other have a correlation null of 60 dB. The point is Audacity is capable of accurate RIAA EQ.
I ran some more tests. At the flip of a switch I can bounce the original RAW playback through the analog RIAA. This adds two conversions to the analog RIAA compare file. In this configuration the analog vs. DSP RIAA have correlation null depths around 10 dB.
I also did simultaneous recording one flat, one with analog EQ. This required two different converters (one USB) with two different footprints. Diffmaker found a 9 ppm error in SR which it had to correct. The analog vs. DSP RIAA has a null depth also around 10 dB.
In both cases the difference sound file revealed three things: (1) An unnulled "blurp" right at the beginning of the file which may affect Diffmaker's null depth measurement. (2) Slight differences in top end which is quite likely converter footprint and (3) minor differences in low end.
The minor difference in low end is not surprising: The DSP version has gain to DC, the Analog version does not have DC gain.
The biquad RIAA available in Audacity using Nyquist or vst are, as a practical matter, as close to analog RIAA as they need to be.
To get a complete picture of the amount of available null, George's mono approach is required. I can use the same converter to acquire both samples. Wire the cart mono. Flat feeds left, right is fed by analog EQ. Only the interchannel A/D differences will affect the difference file.
Other than intellectual curiosity I'm not sure its worth doing because I'm convinced that the two DSP solutions I have for Audacity are probably more accurate than the 1% tolerance analog RIAA network and the Jung/Lipschitz 1% tolerance Inverse RIAA network used to check it.
Sonically the analog and DSP RIAA (if done right) sound virtually identical to me.
EDIT: The difference file comparing the Wayne Stegall Nyquist biquad and analog RIAA (using different convertors) is here: http://www.proaudiodesignforum.com/...ence_File_DSP_Nyquist_RIAA_vs_Analog_RIAA.mp3 (320 kbps mp3 approx 1.2 Mb)
The original files were normalized to -2 dB. I edited the uncorrelated portion at the head out. The peak level of the difference file is about -18 dBFS so the two are very close.
Last edited:
BTW most of my USB devices will not record at 24 bits in Audacity (under Windows) and Cool Edit either post XP. I assume you've followed these discussions here or elsewhere. You simply need to do a histogram test on the noise floor to be sure.
I find some of the portable field recorders are pitched at a higher end audience and are about the same price as a good USB device. That way I get the computer out of the loop and always know what I am getting.
The problems for many users persist at least the developers have stopped saying who needs 24 bits anyway.
Audacity Forum • View topic - 24-Bit Recording Status
I would need to check to see if the Roland Quad Capture is really 24B under XP. I have two PCM4222 eval boards that I don't use often enough which are AES/SPDIF.
I ran Spek on the difference file and most of the differences are in the 10-20 kHz octave. FWIW the analog RIAA is a "passive" and inverting topology; the Nyquist DSP RIAA had no 3.18 µs TC which gives it a slight rising response. (See: Website of Wayne Stegall - Digital Phono Equalization)
I still think the HF differences are converters. The Roland QC has a lot of even-order distortion beginning above -10 dBFS.
Despite the converters affecting measurement, the differences between analog RIAA and DSP-based IIR RIAA are still quite small:
Spectrum of Analog vs. DSP RIAA EQ Difference File
I ran Spek on the difference file and most of the differences are in the 10-20 kHz octave. FWIW the analog RIAA is a "passive" and inverting topology; the Nyquist DSP RIAA had no 3.18 µs TC which gives it a slight rising response. (See: Website of Wayne Stegall - Digital Phono Equalization)
I still think the HF differences are converters. The Roland QC has a lot of even-order distortion beginning above -10 dBFS.
Despite the converters affecting measurement, the differences between analog RIAA and DSP-based IIR RIAA are still quite small:
An externally hosted image should be here but it was not working when we last tested it.
-Logical_Song_Difference File_DSP_Nyquist_RIAA_vs_Analog_RIAA.png){kind=link}
Spectrum of Analog vs. DSP RIAA EQ Difference File
and I need to clarify that it is CoolEdit & Audactiy FIR filters which are Linear Phase & truly EVIL
Linear Phase is the most naive method of doing FIRs. There are MANY DAWs and EQs that are EVIL Linear Phase.
You have to be a analogue filter guru to know that Minimum Phase is Phase as God intended and has loadsa practical advantages besides sounding best.
'Naive' IIRs will almost certain use bi-linear transform, with or without frequency warping .. cos that's Digital Filters 101. The common implementations are usually sorta Minimum Phase.
Many portable recorders are pretty poor and there are zillion bit zillion GHz devices which have the practical performance of a Dolby cassette deck.Scott Wurcer said:
There's a page on one of the Nature Recording sites on these.
The Sound Devices is about the cheapest (??) with near SOTA NF on their mike preamps.
We recommend TASCAM DR680 for the inexpensive recorders but earlier versions had wonky & very noisy P48V and even the latest (improved) version is a long way from 3dB NF on their mike preamps.
Dunno about their line inputs. Their anti-aliasing filters are also often iffy.
If you find a good, inexpensive portable recorder with textbook 16b digital recording, 4 good mike inputs & matched gain, please let us know.
Richard, I think the site you are thinking of for comparison of the input noise of portable recorders is Avisoft Bioacoustics - Microphone Input Noise Comparison
I've been using the Zoom F8 (and recently the F4) and I think they are about as good as the Sound Devices machines at about 1/4 the price. I don't want to derail this topic with more info, those interested can find it.
I've been using the Zoom F8 (and recently the F4) and I think they are about as good as the Sound Devices machines at about 1/4 the price. I don't want to derail this topic with more info, those interested can find it.
I used the Fostex FR2-LE (second on list) for the sound clips on the Linear Audio site. Jan and SY were both there for the live show and the instant replay. The current Tascam recommendation came from a recording professional, I have no complaints.
Last edited:
Yes that can be confirmed what Cool Edit calls FFT filter is linear phase. Let's not get off on a tangent, the needs of nature recordists are very different. BTW noise figure is a virtually useless concept in audio since matched impedance is rarely an issue.
No one would make a multi-bit "real" 16 bit product now except as a fashion statement, it would probably have no anti-aliasing at all as an added feature.
Last edited:
I decided to try a different test approach inspired in part by George's suggestion to wire the cart mono.
The analog RIAA circuit is line level. What I did was simply bridge both channels and fed them line-level pink noise. One output had RIAA EQ applied in analog the other output was RAW (flat). These outputs fed the L and R inputs of the same converter. This eliminates the D/A differences and reduces the A/D differences to the interchannel differences of the convertor which are assumed to be small.
The Analog EQ'd channel was saved as a mono file normalized to -2 dBFS. The RAW file had Olaf Matthes' vst IIR RIAA applied and was saved as a mono file normalized to -2 dBFS.
The two files are identical length, have the same SR and timebase errors.
From these I generated two additional files. One had analog Left, DSP right. The second file was a difference file I created using Cool Edit's paste mix function summing an inverted version of one file with the other.
In an ideal world they should be identical if the RIAA IIR EQ and analog are equivalent. But are they?
Yes. The spectrum of the difference file looks like this:
Analog RIAA vs. IIR-RIAA Difference File Spectrum
Other than low frequencies there is no significant difference between the Analog RIAA and DSP IIR-RIAA.
At low frequencies there is some difference: The analog RIAA does not have DC gain. So what's going on in the lower octaves?
The following persistent goniometer plot (averaged over about 45 seconds) shows the LF phase error of the Analog EQ.
Analog RIAA vs. IIR-RIAA Phase Error
With a 200 Hz HPF applied the phase error disappears:
Analog RIAA vs. IIR-RIAA Phase Error WIth 200 Hz HPF. Note the change in scale.
For all practical purposes the results are identical.
A properly designed IIR RIAA can be just as accurate as an analog RIAA network.
If there's phase error above 100-200 Hz its insignificant. Below 200 Hz the analog RIAA's error is due to desirable infrasonic rolloff.
The analog RIAA circuit is line level. What I did was simply bridge both channels and fed them line-level pink noise. One output had RIAA EQ applied in analog the other output was RAW (flat). These outputs fed the L and R inputs of the same converter. This eliminates the D/A differences and reduces the A/D differences to the interchannel differences of the convertor which are assumed to be small.
The Analog EQ'd channel was saved as a mono file normalized to -2 dBFS. The RAW file had Olaf Matthes' vst IIR RIAA applied and was saved as a mono file normalized to -2 dBFS.
The two files are identical length, have the same SR and timebase errors.
From these I generated two additional files. One had analog Left, DSP right. The second file was a difference file I created using Cool Edit's paste mix function summing an inverted version of one file with the other.
In an ideal world they should be identical if the RIAA IIR EQ and analog are equivalent. But are they?
Yes. The spectrum of the difference file looks like this:
Analog RIAA vs. IIR-RIAA Difference File Spectrum
Other than low frequencies there is no significant difference between the Analog RIAA and DSP IIR-RIAA.
At low frequencies there is some difference: The analog RIAA does not have DC gain. So what's going on in the lower octaves?
The following persistent goniometer plot (averaged over about 45 seconds) shows the LF phase error of the Analog EQ.
Analog RIAA vs. IIR-RIAA Phase Error
With a 200 Hz HPF applied the phase error disappears:
Analog RIAA vs. IIR-RIAA Phase Error WIth 200 Hz HPF. Note the change in scale.
For all practical purposes the results are identical.
A properly designed IIR RIAA can be just as accurate as an analog RIAA network.
If there's phase error above 100-200 Hz its insignificant. Below 200 Hz the analog RIAA's error is due to desirable infrasonic rolloff.
Last edited:
But this is stating the obvious. I know it feels more satisfying to use real signals but it can be proven mathematically. Once you have properly sampled the audio there is no other possibility.
I think the tests prove it.
And they also prove this particular vst plugin (Olaf Matthes) is reasonably (if not exactly) mathematically accurate.
Since Olaf's plugin nulls almost exactly with Wayne Stegall's Nyquist scripts they would seem to be accurate too.
Olaf's (free) RIAA vst plugin is here: http://www.nullmedium.de/dev/audioplugins/
Wayne Stegalls' site where you can build your own Nyquist RIAA coefficients is here: http://waynestegall.com/audio/riaaiir.htm
You can find some ready-to-use Audacity Nyquist RIAA scripts here: http://www.proaudiodesignforum.com/forum/php/viewtopic.php?f=15&t=885
And the link to Scott's RIAA coefficients: https://linearaudio.net/sites/linearaudio.net/files/v10 sw app1 table a-1.xlsx and https://linearaudio.net/sites/linearaudio.net/files/v10 sw Appendix 1b web.docx
Yeah but that will make my head explode.
Last edited:
For the sake of exorcising the fear of the audiophiles toward DSP filters, it’s OK.
In a world of no biases, the phrase would have to be like:
"A properly designed and built analog RIAA network can be just as accurate as an IIR RIAA DSP filter."
George
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.