I’ve been experimenting with digital RIAA equalization over the last couple of days and after many trial and error and having read a lot about the subject, I brew my own setup which is a combination of available equipment and software, and a mix of knowledge I gathered together on this forum along with bits and parts of what I found on the web.
(Thanks to everyone for sharing so much valuable info all over this mighty cyberspace)
The basic idea was to minimize the analogue signal path as much as possible and do everything in the digital domain.
I know a little about analogue and very little about digital which will probably reflect in this post but nevertheless, I achieved ‘something’.
Having a MC cartridge that likes a 47k input resistor on the preamp, it was not a good idea to plug it directly into the M-audio Firewire 410 instrument input which has something like 3k input impedance although that would have been the shortest possible path.
So, I took my beloved Ono clone, consisting of the MC and the MM part and I separated both sections. The reason I kept the MC part was
1) I needed the 47k input impedance and
2) I could use the gain of that stage because, even if the gain of the M-audio instrument input is fairly high, it couldn’t provide enough output with a low enough S/N ratio.
I recorded this non RIAA-equalized signal with Audition and applied an fft filter with the inverse RIAA parameters.
Before that, I removed the ticks and pops and afterwards, I amplified the signal by something like 4dB and finally compared the result with the same recording I previously made with the full fledged Ono.
I have read somewhere that all this is not a very good idea and one of the reasons would be that the low frequencies would be captured with a too low resolution.
Because the RIAA curve boosts the bass at the lowest regions by about 20dB, one needs to record at least 20dB below 0dBfs without the analogue RIAA equalizer.
I don’t get the bass-and-resolution part very well so I would be pleased if someone could explain a bit on the subject.
In any case, comparing both recordings, I have the impression that the new (to me) technique sounds better. Visually (Waveform display), I can see more dynamics but I’m aware that my eyes could trick my ears.
I’ll upload a small extract of both files somewhere and would be happy if some of you could have a listen and let me know what is most pleasing to the ear.
/Hugo
(Thanks to everyone for sharing so much valuable info all over this mighty cyberspace)
The basic idea was to minimize the analogue signal path as much as possible and do everything in the digital domain.
I know a little about analogue and very little about digital which will probably reflect in this post but nevertheless, I achieved ‘something’.
Having a MC cartridge that likes a 47k input resistor on the preamp, it was not a good idea to plug it directly into the M-audio Firewire 410 instrument input which has something like 3k input impedance although that would have been the shortest possible path.
So, I took my beloved Ono clone, consisting of the MC and the MM part and I separated both sections. The reason I kept the MC part was
1) I needed the 47k input impedance and
2) I could use the gain of that stage because, even if the gain of the M-audio instrument input is fairly high, it couldn’t provide enough output with a low enough S/N ratio.
I recorded this non RIAA-equalized signal with Audition and applied an fft filter with the inverse RIAA parameters.
Before that, I removed the ticks and pops and afterwards, I amplified the signal by something like 4dB and finally compared the result with the same recording I previously made with the full fledged Ono.
I have read somewhere that all this is not a very good idea and one of the reasons would be that the low frequencies would be captured with a too low resolution.
Because the RIAA curve boosts the bass at the lowest regions by about 20dB, one needs to record at least 20dB below 0dBfs without the analogue RIAA equalizer.
I don’t get the bass-and-resolution part very well so I would be pleased if someone could explain a bit on the subject.
In any case, comparing both recordings, I have the impression that the new (to me) technique sounds better. Visually (Waveform display), I can see more dynamics but I’m aware that my eyes could trick my ears.
I’ll upload a small extract of both files somewhere and would be happy if some of you could have a listen and let me know what is most pleasing to the ear.
/Hugo
Link to the file:
http://rapidshare.com/files/112338404/Extract.flac.html
The moderate large (13.376 kB) file is encoded with flac and there’s a 2 sec. silence gap between both pieces. Enjoy.
/Hugo
http://rapidshare.com/files/112338404/Extract.flac.html
The moderate large (13.376 kB) file is encoded with flac and there’s a 2 sec. silence gap between both pieces. Enjoy.
/Hugo
Actually it's 13 megabytes, too big for my dial-up, but I'm very interested in the results. From a pure technical standpoint you shouldn't be able to equal the results from a good analog RIAA section. OTOH, if you record the flat signal with enough bits, say 24, and if you can keep the s/n ratio high enough, I'd think you could get a good enough result to make it hard to tell the difference. I've found you have to be very careful of things that sound better during a short listening trial; often they have some subtle flaw that shows up under long term listening. There's also the issue of how well the two RIAA curves match; this is both harder to do and more critical than one might think. You could easily be hearing a mismatch between the hardware and software equalization.
The recording was indeed 24bit and I'm well aware of the many culprits of critical listening. I'll burn the two versions on different CD's, mix them up and try to find out which one sounds best.
I know the Ono has a very accurate RIAA response, and I checked my diy version to make sure I build it right.
So, if the fft filter is as good as the Ono, the only difference should be the greater resolution due to a smaller signal path.
I have uploaded a VBR MP3Pro file which is only 2.2Mb but notice that much of the fine details are lost:
http://rapidshare.com/files/112410743/Extract.flac.mp3.html
/Hugo
I know the Ono has a very accurate RIAA response, and I checked my diy version to make sure I build it right.
So, if the fft filter is as good as the Ono, the only difference should be the greater resolution due to a smaller signal path.
I have uploaded a VBR MP3Pro file which is only 2.2Mb but notice that much of the fine details are lost:
http://rapidshare.com/files/112410743/Extract.flac.mp3.html
/Hugo
I've listened to the Flac-version, and frankly, i can't hear any big differences.
Admitted, i'm listening via my computer, but with a fairly good amp, and a pair of reasonable speakers.
You are making me curious about your fft. I've tried this trick before with Cool dit pro, ( now called Audition ), but wasn't satisfied.
Are you interested in sharing your fft ?
Best regards
Ebbe
Admitted, i'm listening via my computer, but with a fairly good amp, and a pair of reasonable speakers.
You are making me curious about your fft. I've tried this trick before with Cool dit pro, ( now called Audition ), but wasn't satisfied.
Are you interested in sharing your fft ?
Best regards
Ebbe
Of course and I'm trying to refine it ATM.
When recording with AA and perhaps also Cooledit, the fft range varies with the sample rate.
If you record with a 44.1 sample rate the range is 21.39Hz up to 22.050kHz.
I've inputted about thirty frequency points on the scale with the corresponding RIAA numbers and I'm currently comparing a wide range of frequencies by generating sine waves with a fixed frequency and let the fft do the job. Then, with the amplitude stats, I calculate the difference between the original and the RIAA'ed amplitude. The fft size is 8192 points, which should be pretty accurate. I'll post the RIAA curve if you can't find it here on the forum.
/Hugo
When recording with AA and perhaps also Cooledit, the fft range varies with the sample rate.
If you record with a 44.1 sample rate the range is 21.39Hz up to 22.050kHz.
I've inputted about thirty frequency points on the scale with the corresponding RIAA numbers and I'm currently comparing a wide range of frequencies by generating sine waves with a fixed frequency and let the fft do the job. Then, with the amplitude stats, I calculate the difference between the original and the RIAA'ed amplitude. The fft size is 8192 points, which should be pretty accurate. I'll post the RIAA curve if you can't find it here on the forum.
/Hugo
Attachments
The "Why doing RIAA digitally is a bad idea" argument is all to do with dynamic range and easily explained. Draw yourself a graph of dynamic range (in dBs) against log of frequency, then add the RIAA curve at the top with the 50Hz peak just touching 0dBFS (dB Full Scale). Now draw a straight horizontal line at -110dB to correspond to the noise floor of a good practical ADC. Now look at the distance between the RIAA curve and the ADC noise floor at various frequencies; this is your new dynamic range. Finally, consider what happens with variable recording levels, cartridge sensitivities and clicks and bangs. Now you know why people don't usually like to do RIAA digitally.
Still, it's an interesting experiment...
Still, it's an interesting experiment...
Fwiw i didn't like the sound of either segment too much. Still, would prefer the first one for low level resolution and ambience. The second sounds a bit warmer and distortion may well be higher.
Listened through a pair of Leak TL12.1 which i just finished restoring. My normal amp will likely make the difference even more noticeable.
Listened through a pair of Leak TL12.1 which i just finished restoring. My normal amp will likely make the difference even more noticeable.
Thanks for taking the time.
I will not yet reveal who is who as to give others a chance to listen unbiased.
I'm slowly getting a grip on the dynamic range issue.
I statistically compare both files and the digitally eq'd one has a used dynamic range of 32.95dB, while the analogue RIAA has 38.65dB.
/Hugo
I will not yet reveal who is who as to give others a chance to listen unbiased.
I'm slowly getting a grip on the dynamic range issue.
I statistically compare both files and the digitally eq'd one has a used dynamic range of 32.95dB, while the analogue RIAA has 38.65dB.
/Hugo
Netlist said:
Two remarks:
First, you should not get such large quantifiable differences (like 39dB vs 33dB) if you take proper care with the actual measurement method, and match precisely the levels of the recording as well as the actual RIAA curves in both implementations (otherwise its all apples versus aircraft carriers). Unless one or both preamp/RIAA implementations contain gross non-linearities any measurable level difference is probably attributable your own measurement errors.
Two: the Audition FFT filter is linear phase. A RIAA equalizer should be minimum phase. That's a gross deviation from the spec. You have to fix that first before you can compare methods.
There's a pre-made RIAA filter in the Voxengo CurveEQ plugin:
http://www.voxengo.com/product/curveeq/
It comes with three flavours of phaseshift.
Werner,
The measurements I did were both done with Audition (see picture). I would assume the results to be consistent.
The best dynamic range is maintained with the Voxengo. Thanks for that link, Werner.
Then comes the Audition build-in Graphic Equalizer but both are very similar in performance.
30.75 dB vs. 30.15dB
The fft filter does a less good job with only 29.7dB.
I didn’t check the accuracy of the different methods yet.
Still, as mentioned, the analogue equalization is far better, say +5.5dB at best and definitely accurate.
However, I need to attenuate the soundcard inputs more when I use the complete Ono preamp, which means that the initial S/N ratio of the recording is higher when using only the MC part. Filtering out soundcard and other noises would be cheating ATM.
/Hugo
The measurements I did were both done with Audition (see picture). I would assume the results to be consistent.
The best dynamic range is maintained with the Voxengo. Thanks for that link, Werner.
Then comes the Audition build-in Graphic Equalizer but both are very similar in performance.
30.75 dB vs. 30.15dB
The fft filter does a less good job with only 29.7dB.
I didn’t check the accuracy of the different methods yet.
Still, as mentioned, the analogue equalization is far better, say +5.5dB at best and definitely accurate.
However, I need to attenuate the soundcard inputs more when I use the complete Ono preamp, which means that the initial S/N ratio of the recording is higher when using only the MC part. Filtering out soundcard and other noises would be cheating ATM.
/Hugo
Attachments
Another thing I noticed is the difference in waveform between both recording methods.
Left you see the digital and right the analogue version.
The first is not as nicely folded out symmetrically, especially the right (top) channel.
What could be the cause of this and how would I have to interpret this asymmetry?
Left you see the digital and right the analogue version.
The first is not as nicely folded out symmetrically, especially the right (top) channel.
What could be the cause of this and how would I have to interpret this asymmetry?
Attachments
Netlist said:
Perhaps then your assumption is wrong. You are using very crude statistical methods on datasets which are very very large. The Audition stats work in time slices, glossing over finer detail in the waveform/dataset to get quickly to reasonably-accurate results. If you don't align the datasets with single-sample accuracy then you can't compare, or at least you can't use the Stats tools as the measurement tool you want it to be.
Further you really have to normalise the datasets with respect to average level in the midband (say 300-1000Hz) and perhaps even below 20Hz: LP replay causes such a heap of subsonic rubbish that any LF 'difference' in the signal chain (I guess the Ono is AC coupled, and the path use used for digital was near-DC) may warp the Stats results.
Netlist said:
Audition's waveform displayer is even cruder. Again, you can't extract any meaningful information from this unless you zoom in to single sample accuracy.
Trust me, I've been using Audition for lab purposes for many years now.
A couple of comments. Why not split the difference? You probably don't want an input pre-amp with enormous BW anyway, so why not do the last 20dB with a simple analog roll-off at the third RIAA time constant? 24/96 should be able to take care of the rest of the dynamic range.
If you want to play with filters, running them over an ideal step or impulse can give a clearer picture. It's unfortunate that there isn't a switch in Audition to force minimum phase behavior. It's easily available in the SDK, but that's too much like work.
My experience with LP transfers really hinged on the crest factor of the piece and how much of the A/D range I could use, Even to the point of doing dry runs.
Another point, when I only had a 16 bit recorder I noticed that while monitoring the spectrum much of it (at the high end) remained below the 96dB or so noise floor. Also on some LP's the noise of the master recorder being turned on and off is clearly visible well above even the 16bit level. I have some field recordings from the 70's made on a Nagra by an experienced phonographer that show this easily. He tried to keep everything simple and processing free.
If you want to play with filters, running them over an ideal step or impulse can give a clearer picture. It's unfortunate that there isn't a switch in Audition to force minimum phase behavior. It's easily available in the SDK, but that's too much like work.
My experience with LP transfers really hinged on the crest factor of the piece and how much of the A/D range I could use, Even to the point of doing dry runs.
Another point, when I only had a 16 bit recorder I noticed that while monitoring the spectrum much of it (at the high end) remained below the 96dB or so noise floor. Also on some LP's the noise of the master recorder being turned on and off is clearly visible well above even the 16bit level. I have some field recordings from the 70's made on a Nagra by an experienced phonographer that show this easily. He tried to keep everything simple and processing free.
Werner,
Point taken but the stats analyzing tool really only starts to work from samples about half a second of music and bigger. That’s quite a lot of samples, certainly in 32bit mode. I believe this tool is mainly designed to get average statistics of a full track
Anyway, I realize that apples should be compared with apples and one of the main problems is to adjust the input sensitivity of the soundcard every time I switch from Ono to ‘Half Ono’ mode.
I really have a hard time to understand how digital works so all input is highly appreciated.
Scott,
What is BW?
As for the filters, I read that the advantage of fft is a very low noise floor, far below the soundcard noise so it wouldn’t add up to the music. Unfortunately, something nasty is going on with the phase.
The Voxengo plugin is doing a real good job and is extremely accurate. 0.01db above +/- 140Hz.. Below, something like 0.3dB at 20Hz but even that can be adjusted.
The only two problems I’m still facing is the asymmetry of the waveform when recording straight from the cartridge (Only MC amp without RIAA) and the higher noise floor of the instrument input of the soundcard.
Perhaps, in a later experiment, I might isolate the cartridge with a transformer if such things exist.
And yes, when looking at spectrum of music, one can see all odd things like straight lines in the 16kHz region (bad filtered out oscillators I presume), sometimes the splicing of the tape and small drop-outs. Hotel California has a nice one.
/Hugo
Point taken but the stats analyzing tool really only starts to work from samples about half a second of music and bigger. That’s quite a lot of samples, certainly in 32bit mode. I believe this tool is mainly designed to get average statistics of a full track
Anyway, I realize that apples should be compared with apples and one of the main problems is to adjust the input sensitivity of the soundcard every time I switch from Ono to ‘Half Ono’ mode.
I really have a hard time to understand how digital works so all input is highly appreciated.
Scott,
What is BW?
As for the filters, I read that the advantage of fft is a very low noise floor, far below the soundcard noise so it wouldn’t add up to the music. Unfortunately, something nasty is going on with the phase.
The Voxengo plugin is doing a real good job and is extremely accurate. 0.01db above +/- 140Hz.. Below, something like 0.3dB at 20Hz but even that can be adjusted.
The only two problems I’m still facing is the asymmetry of the waveform when recording straight from the cartridge (Only MC amp without RIAA) and the higher noise floor of the instrument input of the soundcard.
Perhaps, in a later experiment, I might isolate the cartridge with a transformer if such things exist.
And yes, when looking at spectrum of music, one can see all odd things like straight lines in the 16kHz region (bad filtered out oscillators I presume), sometimes the splicing of the tape and small drop-outs. Hotel California has a nice one.
/Hugo
I’ve played quite a bit with your suggestion and it is definitely nice to see that the dynamic range can be doubled, or almost.
The trick is to adjust the software filter so that the final waveform corresponds with the theoretical RIAA curve. Not easy but once set, it should stay there.
A nice feature of 32-bit recording is that one doesn’t have to care about the 0dBFS limit anymore. As long as the soundcard input doesn’t clip, the software takes the bits, undistorted.
Care to elaborate a bit more on the crest factor and how to resolve it?
/Hugo
The trick is to adjust the software filter so that the final waveform corresponds with the theoretical RIAA curve. Not easy but once set, it should stay there.
A nice feature of 32-bit recording is that one doesn’t have to care about the 0dBFS limit anymore. As long as the soundcard input doesn’t clip, the software takes the bits, undistorted.
Care to elaborate a bit more on the crest factor and how to resolve it?
/Hugo
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.