Sorry to be pedantic, I can't help it. The point simply is that the continuous amplitude vs. frequency behavior of an IIR filter is completely determined by the coefficients and the filter does not need to be applied to data to derive it.
I recall a story in which Edison asked some of his mathematical geniuses what the volume of a particular bulb shape was. They spent a few days solving the apparently complex problem and when Edison returned to the lab asked them what the results were. They went on to explain their methodology in great professorial detail and told him what they thought the volume was.
Edison is reported to saying something like "Let's see if you're right." He took the glass model they were using for measurements, filled it full of water and dumped the fluid into a graduated beaker. Edison pointed to the beaker and is reported to have said: "That gentleman is your answer."
Exactly.
And I wanted to make sure the tools that I found were actually accurate and didn't want to have to take the developer's word for it.
Edison is reported to saying something like "Let's see if you're right." He took the glass model they were using for measurements, filled it full of water and dumped the fluid into a graduated beaker. Edison pointed to the beaker and is reported to have said: "That gentleman is your answer."
Exactly.
And I wanted to make sure the tools that I found were actually accurate and didn't want to have to take the developer's word for it.
Last edited:
We had a similar problem when we made our first 16 bit DSP multiplier, how do you test it? Try all 2**32 possible inputs and check each one?
media (name?),
Having a testing attitude is great and your contribution is very welcome.
My
comment was on that particular phrase.This would be nice to have it explained (what signal does exercise all bits and how)
George
It's Wayne Kirkwood. I don't have a horse in the DSP or analog race. I can have it either way at the flip of a switch. No bias intended.
My repair workflow uses analog RIAA EQ primarily for monitoring so its good to know that above 100 Hz or so I can count on the analog to sound like the DSP-EQ'd file. My speakers likely appreciate the dB or so added rolloff at 20 Hz.
Now that I proved to myself that the IIR DSP tools I have available are not some FIR solution posing as RIAA EQ I'm more comfortable rendering an entire library.
One of my objectives is for people to have ready-to-use plugins that don't require advanced math, rocket science, membership in a secret club or searching more than about 60 seconds on Bing. The links are here:
Olaf's (free) RIAA vst plugin is here: Audio Plug-Ins
Wayne Stegalls' site where you can build your own Nyquist RIAA coefficients is here: Website of Wayne Stegall - Digital Phono Equalization
You can find some ready-to-use Audacity Nyquist RIAA scripts here: http://www.proaudiodesignforum.com/f...php?f=15&t=885
And the link to Scott's RIAA coefficients: https://linearaudio.net/sites/linear...ble a-1.xlsx and https://linearaudio.net/sites/linear...01b web.docx
My repair workflow uses analog RIAA EQ primarily for monitoring so its good to know that above 100 Hz or so I can count on the analog to sound like the DSP-EQ'd file. My speakers likely appreciate the dB or so added rolloff at 20 Hz.
Now that I proved to myself that the IIR DSP tools I have available are not some FIR solution posing as RIAA EQ I'm more comfortable rendering an entire library.
One of my objectives is for people to have ready-to-use plugins that don't require advanced math, rocket science, membership in a secret club or searching more than about 60 seconds on Bing. The links are here:
Olaf's (free) RIAA vst plugin is here: Audio Plug-Ins
Wayne Stegalls' site where you can build your own Nyquist RIAA coefficients is here: Website of Wayne Stegall - Digital Phono Equalization
You can find some ready-to-use Audacity Nyquist RIAA scripts here: http://www.proaudiodesignforum.com/f...php?f=15&t=885
And the link to Scott's RIAA coefficients: https://linearaudio.net/sites/linear...ble a-1.xlsx and https://linearaudio.net/sites/linear...01b web.docx
You may also be interested in this balanced vs. unbalanced noise floor comparison.
The source is a Stanton 681EE with the arm at rest.
The cartridge is wired floating with shielded twisted pair.
The input stage is an instrumentation amp input with a dual cross-coupled common mode rejection stage.
RIAA EQ is applied, no A-weighting.
There was an operating light dimmer within 5 feet of the cartridge.
The green trace is the left channel of the preamp when it is balanced.
The blue trace is with the unbalancing jumper (at the input) installed.
Everything else is completely identical.
Note that the absolute dBu calibration is not accurate - this is a relative measurement between a balanced and unbalanced connection.
Wiring a cart/turntable/arm fully-balanced is silly easy. The majority are already balanced up to the arm base.
Despite the theoretical disadvantage of of an "extra" op amp in the real world I'll use the balanced connection every time. YMMV.
The source is a Stanton 681EE with the arm at rest.
The cartridge is wired floating with shielded twisted pair.
The input stage is an instrumentation amp input with a dual cross-coupled common mode rejection stage.
RIAA EQ is applied, no A-weighting.
There was an operating light dimmer within 5 feet of the cartridge.
The green trace is the left channel of the preamp when it is balanced.
The blue trace is with the unbalancing jumper (at the input) installed.
Everything else is completely identical.
Note that the absolute dBu calibration is not accurate - this is a relative measurement between a balanced and unbalanced connection.
Wiring a cart/turntable/arm fully-balanced is silly easy. The majority are already balanced up to the arm base.
Despite the theoretical disadvantage of of an "extra" op amp in the real world I'll use the balanced connection every time. YMMV.
I suppose Olaf uses Robert Orban's fitted coefficients ... if this is the situation then results are excellent otherwise but as there's tiny error in the lower frequency area because of he used 50.5Hz instead of 50.05Hz in his coefficient calculations....
Stegall uses MZT in his realization (BLT is more common) and therefore has some trouble at near Nyqvist frequency (this can be improved by either raising the sampling frequency (384kHz gives quite good results) or using some fitting method for frequency response optmization).
With Audacity you usually use FIR filters which is not good type for RIAA EQ.
Here's my latest digital RIAA software you might want to try out (Max/MSP runtime needed https://cycling74.com/downloads ).
Last edited:
Not sure I understand the question Scott.
Yes I know. I did try Cool Edit's FIR initially and it was bizarro world. Knowing that Audacity's native filters were FIR I didn't even go there. Until a couple of days ago I had never even ventured into Audacity's FIR RIAA menu.
I use the IIR plugins (or Nyquist script) and started with yours jiiteepee. Any idea why it won't run under Win 7 x64 running 32B Audacity? XP SP3 32B runs fine. And what happened to Dave Haupt? His site has a link to his sleep study.
I think Audacity has an undeserved bad reputation for RIAA when it shouldn't. Just don't use its native FIR EQ and all's well.
Thanks for posting the link to your new RIAA EQ. I'll give it a try.
Last edited:
The noise floor if input limited should follow the RIAA -20dB from 1k to 20K. It is possible that the sound card does not have enough gain in front of it or the cart + termination + RIAA combines for this result.
Minimum phase FIR's give exactly the same result as the IIR's (in fact better).
Bob Orban does deserve credit for the derivation of the best possible fit with IIR's but his stuff did make my head explode. Luckily it was not that hard to use some simpler optimization techniques to find the global minimum solution.
At high sampling rates IIR's 101 works fine the tweeks become less and less necessary.
Robert Orban wrote:
>
>> I can't see the first part of this thread, but if you are trying
>> to do an IIR simulation of the RIAA phono de-emphasis curve
>> (assuming s-plane poles at 50.5 and 2122 Hz and an s-plane zero
>> at 5005. Hz), here are some good minimum-phase magnitude
>> approximations.
>
>Neat, thanks! How did you make them?
I used a program I wrote (in ye olde Fortran
follows:
Given a desired magnitude response in the z-plane, there exists a
response in a frequency-warped u-plane that, when bilinear-transformed
to the z-plane, creates the desired z-plane magnitude response.
-Compute the [magnitude response]^2 of the s-plane prototype on a grid.
This is the square of the desired z-plane response.
-Warp the frequency axis by using the bilinear transform, recognizing
that we are approximating using omega^2 as our frequency variable. The
warp maps Nyquist to infinity.
-Make a least-squares rational approximation (i.e., ratio of
polynomials) to the values on the frequency grid. (I used the Numerical
Recipes routine RATLSQ, which uses Chebychev polynomials.)
-Refine the approximation to make the fractional error minimax by using
Remez's Second Algorithm [which applies to rational functions; it's not
the same as the Remez algorithm used in the classical MPR FIR design
program; see Forman S. Acton, Numerical Methods That Work (Revised
Edition), Washington D.C., American Mathematical Society, 1990, pp 310-
314]
-Transform the result into the z-plane in two steps. The first
recognizes that we have been approximating using the magnitude square
function, so we must take the square roots of the poles and zeros of the
approximated rational function, taking the negative real parts to
guarantee a stable, minimum-phase function. The second step is to apply
the bilinear transform to the result of the first step. This yields the
final z-plane poles and zeros.
There are some "interesting" numerical issues in making this procedure
work, mainly because the Remez update formulas require solving a system
of mildly nonlinear equations that tend be ill-conditioned.
The nice thing about the algorithm is that the frequency-warping moves
Nyquist to infinity and thus increases the resolution of the
approximation close to Nyquist, which is where difficulties often occur.
Robert Orban
Last edited:
It is both things primarily the later.
The gain is a little low since its optimized for flat (non-EQ'd) peak levels and 24 dB HR at 1 kHz. Although the RIAA gain can be independently adjusted I have it near unity gain (@1 kHz) to reduce apparent level changes when monitoring pre/post EQ.
The primary reason the noise floor does not follow the EQ curve can be seen in the noise floor response of the flat preamp where the cartridge inductance causes the noise floor to rise in opposition to the RIAA curve falling.
The flat preamp output noise floor with an OPA2134 and balanced input:
With a resistive input termination and more gain you'd likely see the noise floor more closely follow the RIAA curve.
Last edited:
As all this discussion is making me actually want to try ripping as well as feeding the miniDSP what is the starting point for a USB ADC that's good enough? I have to decomission the PC in the listening room so I can only use laptop solutions. Lovely though the benchmark ADC-1 is, not something I will be able to afford or justify.
The Scarlett 2i2 works fine the Xonar's at <$100 would surprise you (maybe). Hard for me to recommend on ears only, I figure we need to uphold our own standards. I had an RIAA that was +5dB off over more than a decade and for the life of me I couldn't really say it was audible. Everyone knows I can't hear anything but noise anyway.
I came across a very thoughtful blog by someone IIRC from Toronto who was a member of a very serious ($$$$) group of listeners. The discussions on how good the whole system, room, equipment, ambient noise level, etc. have to be to even begin to hear these so called "obvious" differences were eye opening.
Well when I was keeping a PC there a Xonar PCI-E was top of my list, so would not be that suprised. TBH the question was more around the horror stories mentioned about not actually being able to record at 24/96 and confirming that there aren't some horrors out there with 24 bit ADCs and 14 bit noise floors.
The Scarlett appears to fit the bill
There are components which are not (fully) x64 compatible so maybe it's the reason. IIRC, DH has been out of business quite long time already ... can't remember the reason for that but fortunately his addons can still be found.
Well when I was keeping a PC there a Xonar PCI-E was top of my list, so would not be that suprised. TBH the question was more around the horror stories mentioned about not actually being able to record at 24/96 and confirming that there aren't some horrors out there with 24 bit ADCs and 14 bit noise floors.
The Scarlett appears to fit the bill
IME you will need something like Reaper, JRiver, or use Linux to get 24/96 for sure.
I did wonder about using arecord from the command line on my music server, but that seemed a bit masochistic. Sadly a headless unit and as it works in its intended role don't want to rebuild it at the moment. The 2i2 comes with a cut down protools which would be disappointing if it didn't allow saving at 24/96.
The TI PCM4222EVM Evaluation Module is a very good A/D value. ($149 US) http://www.ti.com/lit/pdf/sbau124
The PCM4222 EVM does not have USB connectivity but it does have transformer-isolated AES, transformer-isolated SPDIF and an audio serial port. The PCM4222EVM is a bare PCB, requires an external power supply and case but is well worth the added DIY effort. Does 192K 24B. I haven't researched it but you may find an off-the-shelf USB transceiver that will hook right up. The audio input is balanced low-impedance XLR/TRS combo connectors. I use mine with a Meanwell 25W desktop (brick) switcher to provide +/-15 and 5V.
I have two PCM4222EVM modules and have used them for instrumentation. The input connector feeds the OPA1632 modulator driver and then the PCM4222 so there are no intervening PGAs or mic/instrument preamps made to look like line inputs to add noise and distortion etc. The board is very clean and it showcases the PCM4222 quite well.
What I don't recommend is a Roland Quad Capture. It's OK for MI. I bought it on recommendation from SpectraPlus and the Quad capture turned out to be useless as a bench interface. The shortest signal path from line input to USB begins to overload around -10 dBFS. A longer path, through its mixer, can provide a higher overload point but the resulting gain increases the noise floor. It's DAC output is equally useless having high even-order THD.
With USB devices I've always had to fight 1 kHz spurs from USB polling and it's difficult to obtain galvanic isolation.
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.