Forgive me for asking a specific question which perhaps has been dealt with in these 30-odd pages ... I read the first few - and the last few - and couldn't get the answer (but can't face trawling through the remaining pages).
I am about to start digitising the output of my phono stage, so I can feed it into the digital input of my miniDSP 10x10Hd.
The ADC I'll be using is 24/192 and has an input maximum of 2v.
I can adjust the gain of my phono stage to vary the output level between 200mV and 600mV - so my question is ... what is the optimum signal level (average) to feed into the ADC?
If I set the phono stage gain so that a 4mV MM delivers 200mV into the ADC ... then I am allowing 20dB (10x) headroom for peaks. However, because most of the time I am feeding only 200mV into the ADC ... I am not as far above the noise floor as I would be if I were feeding an average of, say, 500mV into the ADC.
Yet if I (on average) feed a 500mV signal into the ADC ... I only have headroom of 4x.
So we have to balance available headroom vs. distance above the noise floor - has anyone recommended the optimum?
Thanks,
Andy
I am about to start digitising the output of my phono stage, so I can feed it into the digital input of my miniDSP 10x10Hd.
The ADC I'll be using is 24/192 and has an input maximum of 2v.
I can adjust the gain of my phono stage to vary the output level between 200mV and 600mV - so my question is ... what is the optimum signal level (average) to feed into the ADC?
If I set the phono stage gain so that a 4mV MM delivers 200mV into the ADC ... then I am allowing 20dB (10x) headroom for peaks. However, because most of the time I am feeding only 200mV into the ADC ... I am not as far above the noise floor as I would be if I were feeding an average of, say, 500mV into the ADC.
Yet if I (on average) feed a 500mV signal into the ADC ... I only have headroom of 4x.
So we have to balance available headroom vs. distance above the noise floor - has anyone recommended the optimum?
Thanks,
Andy
Hi Andy
I go for 20dB (x10) headroom and I don't have to look back.
Bill thanks.
Here is the second part
Phono Stages for Digital Ripping
George
I go for 20dB (x10) headroom and I don't have to look back.
Bill thanks.
Here is the second part
Phono Stages for Digital Ripping
George
Thanks, George - so with 20dB headroom, you don't get a noisy result, when listening to the digital rip?
Andy
George: You beat me to it I was re-reading Scott's LA paper to see what his advice was. I came up with 24dB so we are close enough. I blame the fact that I have quite a few recordings that are a tad hot. But as my cartridge selections cover 2.5mV to 5mV output I need some gain adjustment anyway so a bit more range on that makes sense.
FWIW and in case I have the sums wrong I have assumed that real music has a roughly pink noise distribution so the boost above 2kHz doesn't have to be accounted for in ADC headroom. Hottest records you can find are +18dB and (from Scott) allow 6dB for the worst clicks n pops assuming as he found that ADCs clip cleanly.
I must have checked tubecad the day before he put the new article up. He really does have an amazing workrate 🙂
FWIW and in case I have the sums wrong I have assumed that real music has a roughly pink noise distribution so the boost above 2kHz doesn't have to be accounted for in ADC headroom. Hottest records you can find are +18dB and (from Scott) allow 6dB for the worst clicks n pops assuming as he found that ADCs clip cleanly.
I must have checked tubecad the day before he put the new article up. He really does have an amazing workrate 🙂
The circuits that always have 47K in series with the input are a bad idea noisewise.
I don’t think it makes much difference on a 24 bit recording.
Scott, I guess you’ve read the text beneath the JLH schematic.
I am tempted to build both the inverting and the non inverting IC op amp versions to judge for myself.
George
Attachments
In order to reduce the thread drift on the 'mechanical resonance in MM' thread I though I would pop a link in here. Richard (Kgrlee) posted a link to some stuff Wayne Kirkwood was working on for MC stages. Pro Audio Design Forum • View topic - A Low Noise Balanced In Moving Coil Preamp Using the ZTX851 . I am still digesting and notice some sims from HansP in there. But looks like a good solution for those of us with silly low output (.25mV) MCs in our arsenals. I also ought to splash out for a copy of H&H v3 to thank them for finding the ZTX851 🙂
I said I'd move the digital stuff from
http://www.diyaudio.com/forums/analogue-source/303389-mechanical-resonance-mms-59.html#post5050174 here but have been a bit slack cos some beach bum issues over Easter.
Please bear with me as I'm also restoring a wooden sailing boat ... which is self explanatory for anyone who has done similar. The recent cyclones in Queensland have pushed me to work harder on this. 😱
My stuff is mainly in Simple Arbitrary IIRs and I hope to show where this is applicable and where other methods may be preferable.
http://www.diyaudio.com/forums/analogue-source/303389-mechanical-resonance-mms-59.html#post5050174 here but have been a bit slack cos some beach bum issues over Easter.
Please bear with me as I'm also restoring a wooden sailing boat ... which is self explanatory for anyone who has done similar. The recent cyclones in Queensland have pushed me to work harder on this. 😱
My stuff is mainly in Simple Arbitrary IIRs and I hope to show where this is applicable and where other methods may be preferable.
Richard I'll ask again post some worked through examples, starting with the three time constant RIAA. I can find only one article that references your paper, with all due respect this does not necessarily mean anything. There just are no platforms for complex arbitrary feedback IIR's, people want at least some semblance of a cookbook process. 200 IIR coefficients with a multitude of delay feedback/forward paths is not cookbook. Every case must be coded by hand into a DSP, etc.
IME asking folks to pay the $33 will end the conversation, just the way it is.
Last edited:
Please bear with me Scott. I'm trying to find code which went from s=jw to z but it's embedded in stuff which I haven't looked at in nearly 10 yrs.
Can you do me a favour please. Could you send me a copy of your 48k 3120pt RIAA FIR. Also a large eg 32K pt. RIAA impulse preferably in the form of a WAV file.
I mistook your 3120pt filter for 32K and was suitably awed 😱 3120 is within range of stuff I've done before 🙂
I can't process impulses longer than 8K with my 1980s tools. Don't laugh
BTW, I wouldn't use SAI to convert anything for which I already had a s polynomial .. like RIAA. I would use mostly Binomial Transform for most of the frequency range and other stuff at LF .. dealing with Frequency Skewing as required. I admit I've never done Frequency Skewing for real as SAI handles most of my 'HF' stuff.
_________________
The pic below is Fig 4.1 from SAI.
All da platforms which allow you to define an EVIL FIR already has the top half and the accumulator. I assume, they allow you to enter a list of the b0 - bN feedforward coeffs and to define N, the length of the FIR and Fs, the sample rate. The most convenient way to do this is to have the FIR as a WAV file.
To define an Arbitrary IIR, all you need is the bottom half which is another chunk of memory (which almost certainly exists as an output buffer), a similar bit of code to add the feedback to the accumulator and some way to enter A1 - AM.
There are stability issues with IIRs both in design and the resulting filters ... which many of the references in SAI report. I believe SAI avoids most of the problems and is proven to design bigger IIRs than competing methods without problem.
I'm not that keen for loadsa people to understand & use SAI as it is a big edge in the commercial products that currently use it. But I'm certainly amenable to dreaming up the odd filter for Bill or George ... assuming they have a platform that will accept an arbitrary number of Feedforward & Feedback coeffs.
An alternative method .. if all the cheapo digitial platforms have is a bunch of Bi-Quads is
[25] Ramos G & Lopez J, "Filter Design Method for Loudspeaker Equalization Based on IIR Parametric Filters", JAES 54 no12 dec06
Ramos & Lopez [25] is a robust, but computationally intensive alternative. They automate an obvious method to use a large number of parametric EQs.
The largest discrepancy between the current response and the target is EQd with the first parametric, then the largest remaining discrepancy with the second parametric … and so on. When the parametrics are used up, go back to the first parametric and see if it can be adjusted for further improvements … and so on. Continue iterating until some criteria is met.
You can certainly do this manually and get good results if you have the time to twiddle.
Attachments
Du.uuh! If you are using Guru Wurcer's EVIL 3120pt FIR for RIAA, it has easily enough power to correct cartridge mechanical & electrical effects within the same filter.
Just have to watch the measurement caveats in SAI and make sure you have Min. Phase .. but no need to listen to beach bums pontificating as pseudo DSP gurus 😀
Could be done in MATLAB or the freeby Gnu Octave without hi-falutin maths ..
Or you could do RIAA with Guru Wurcer's 2 bi-quads and the cartridge EQ with an FIR
Last edited:
Betting on even a short term freeze on computational power is a losing proposition. In 6mo or a year the brute force open source techniques are adequate again.
There aren't any there must be a reason for that. I suspect it is very easy to get instability or numerical noise issues. Even the order of two bi-quads has problems at single precision FP.
Last edited:
Even beach bums are allowed vanity 🙂
I did this circa 1990 but at the time, practical implementation was difficult & expensive.
In da 21st century, implementation (for the user) is still difficult but no longer expensive.
It only takes ... eg the arbitrary IIR form I describe to become common in a cheapo DSP platform .. or even in EVIL obfuscating Windoze audio implementations .. for it to become simple.
From your comments, the paper is probably not useful for you.
I'm betting that computational power is going to make arbitrary IIRs easy on cheapo handheld devices like phones & tablets 😀Scott Wurcer said:
Precisely. Anyone playing with 'big' IIRs soon encounters these. The references I quote are quite illuminating. SAI addresses many of these problems .. including some of the problems at the frequency extremes.
Scott, thanks for pointing out the 'frequency warping' caveats. I should make more of a point about them.
Bill & George, I think using Guru Wurcer's 3120 pt RIAA to incorporate the cartridge stuff is the quickest & easiest way forward.
If you can produce a impulse response of what you want corrected I can 'modify' Scott's 3120 pt FIR to incorporate correction while alleviating some of the FIR EVILs. 🙂
We can discuss how to get this impulse (or other) response.
First we need our trustworthy Test Record.
_____________________
How does the MiniDSP accept a large FIR? Is it just a text list of coeffs?
I'm assuming it does an FIR with the Direct Form I filter I show in #332. But there's a distinct possibility it uses a FFT block scheme for FIRs as large as 3120.
Won't affect what I suggest in this post ... but it might affect simply implementing an arbitrary IIR 😡
If you can produce a impulse response of what you want corrected I can 'modify' Scott's 3120 pt FIR to incorporate correction while alleviating some of the FIR EVILs. 🙂
We can discuss how to get this impulse (or other) response.
First we need our trustworthy Test Record.
_____________________
How does the MiniDSP accept a large FIR? Is it just a text list of coeffs?
I'm assuming it does an FIR with the Direct Form I filter I show in #332. But there's a distinct possibility it uses a FFT block scheme for FIRs as large as 3120.
Won't affect what I suggest in this post ... but it might affect simply implementing an arbitrary IIR 😡
Last edited:
Sorry for the exeptional late reply. I never read the thread until now.
Strange how people think digital noise is something completely different than analogue noise, its not. Noise is noise.
Yes, we can hear sounds below the noise level, sometimes even 20 dB below the noise level. But this applies to digital systems just as well.
So your conclusion that we need to target for a signal 20dB below the noise floor of vinyl, is wrong. Information theory is very clear on this: In order to completely transfer a signal with limited bandwidth and a noise level, all you need is a channel that has a tiny bit bigger bandwidth and a tiny bit lower noise floor.
Iow: ANY audio AD/DA converter on the market today, even the cheapest ones you can find, performs better than the best vinyl playback systems out there.
Strange how people think digital noise is something completely different than analogue noise, its not. Noise is noise.
Yes, we can hear sounds below the noise level, sometimes even 20 dB below the noise level. But this applies to digital systems just as well.
So your conclusion that we need to target for a signal 20dB below the noise floor of vinyl, is wrong. Information theory is very clear on this: In order to completely transfer a signal with limited bandwidth and a noise level, all you need is a channel that has a tiny bit bigger bandwidth and a tiny bit lower noise floor.
Iow: ANY audio AD/DA converter on the market today, even the cheapest ones you can find, performs better than the best vinyl playback systems out there.
. Agreed B&K used a saw to make their impulse response!
Mine doesn't IIR or IIR are the options, just bung in the coefficients. The big $$$ miniDSP units with Dirac do, but not going to buy one of those. If I was starting again would look closely at the hypex unit, which is pretty good for the money.
- Status
- Not open for further replies.