The Phonoclone and VSPS PCB Help Desk

I was intrigued by Wirehead's measurements of his VSPS, so I thought I'd try measuring my VSPS using the same method. Here's my process:

  1. In RMAA, generate 96 KHz/32-bit float test signal files
  2. In Audacity, open the test signal and:
    1. Reduce the amplitude by -27.5 dB via the "Amplify" effect to avoid clipping high frequencies in the next step
    2. Apply the inverse RIAA curve by selecting the "Equalization" effect, choosing the "RIAA" curve, clicking the "Invert" button to invert the curve, and increasing the length of the filter to the maximum 8191 for improved accuracy
    3. Export the test signal
  3. Insert the VSPS in a loopback between an output & input of my E-MU 1616M
  4. Playback the test file in foobar2000 and record in Sound Forge, all using ASIO
E-MU's software/hardware mixer PatchMixDSP was configured to send ASIO outputs directly to the hardware outputs, and hardware inputs to the ASIO inputs to avoid any unwanted bit manipulation by the drivers/OS. The 1616M's input reaches full-scale at 2V RMS so I reduced the output signal from the 1616M in PatchMixDSP by 14 dB to get the VSPS to amplify the signal to that level.

The VSPS was configured as follows:

  • DIY PCB
  • 2x OPA827 on a single-to-dual SO8-to-DIP8 BrownDog adapter
  • Vishay CMF55 1% resistors, unmatched, 105k/732k for R4/R5 respectively
  • Panasonic ECQ-P(Z) 1% polypropylene capacitors, unmatched
  • Nichicon Muse ES 10uF bi-polar capacitor for C3 and 100k for R7
  • 2x 9V batteries in series, with a TLE2426 rail splitter for ground
As you can see the measured performance is very, very good. Even with unmatched 1% components the frequency response reflects RJM's simulations damn near perfectly and with exceedingly good matching between channels. As far as I know the Panasonic ECQ-P(Z) capacitors are now discontinued and unavailable from DigiKey (where I purchased them), but I'm sure there are other companies that manufacturer equivalent caps with the same tolerance.

Attached is the full RMAA test in HTML for you to view, and a .sav file you can load up in RMAA. The "roughness" below ~1 KHz is probably due to all of the digital filtering done for the inverse RIAA and level reduction.

Not sure what to think of the slight rise in frequency response below 50 Hz though, even though it's not enough to be much of a bother.

With all that said, thank you RJM for the VSPS and also to Wirehead for tips on how to measure this thing!
 

Attachments

  • VSPS Frequency Response.png
    VSPS Frequency Response.png
    3.1 KB · Views: 198
  • VSPS-RMAA.zip
    247.5 KB · Views: 33

rjm

Member
Joined 2004
Paid Member
Is the Allen Wright modification that is being referred to the same as adding the 50kHz filter?
Is this a pole or a zero?
My reasoning is that it is a zero (= going flat beyond 50kHz).

Yes on both counts.

The zero shows up anyway in a non-inverting opamp circuit, since the gain can be no less than unity. All I did was move the zero up a little to set it at 50 kHz.

@vulejov,

My copy of Horowitz and Hill runs 1125 pages, but you'd need only the first 59, chapter 1 "Fundamentals" explains how impedance and passive filters work and how to calculate the response.

The RIAA is not defined by a circuit, but by three time constants. There are any number of ways it can be realized in practice but the values generated by calculation or software like Audacity are exact by definition.

@L-Train

That's really impressive, I suppose 1% components makes all the difference here, as most people will be only using 5% capacitors. Funny how the bass frequency tips up so very slightly, but that's the result me setting R5 just slightly high. I kinda slipped that in, figuring it will compensate for the rolloff generated by the output coupling cap C3.
 

rjm

Member
Joined 2004
Paid Member
Up for test was the TI NE5532; the BB OPA2134 and the National LM4562NA.

Take a guess as to which one came out best in terms of noise floor/dynamic range. :D

An educated guess, based off the datasheets:

For an input impedance of 470 ohms, 2134 is 8.5 nV/sqrtHz, 5532 works out to 5.8 nV/sqrtHz, while the 4562 slips in at 3.9 nV/sqrtHz. For completeness, the OPA27 is at 4.2 and the LT1115 at 3.1 nV/sqrtHz.

As expected, at these low impedances the trusty 5532 is more than up for the task. The 2134 is a better choice to sit behind a 100k volume pot for example, its my go to model for headphone amps and preamps.

Douglas Self was basically right though: at the end of the day you can't go far wrong with the 5532-5534. Just a really nice, easy to implement balance of low noise, low distortion, good drive, and stability.
 
Last edited:
@vulejov,

My copy of Horowitz and Hill runs 1125 pages, but you'd need only the first 59, chapter 1 "Fundamentals" explains how impedance and passive filters work and how to calculate the response.

The RIAA is not defined by a circuit, but by three time constants. There are any number of ways it can be realized in practice but the values generated by calculation or software like Audacity are exact by definition.

On what basis we are confident that the simulations using the correct equation?

How do we know that the progam is using the correct inverse RIAA?

Except incorrect attempt adding two values ​​so far we have no evidence that the RIAA curve is correct..
 
An educated guess, based off the datasheets:

For an input impedance of 470 ohms, 2134 is 8.5 nV/sqrtHz, 5532 works out to 5.8 nV/sqrtHz, while the 4562 slips in at 3.9 nV/sqrtHz. For completeness, the OPA27 is at 4.2 and the LT1115 at 3.1 nV/sqrtHz.

As expected, at these low impedances the trusty 5532 is more than up for the task. The 2134 is a better choice to sit behind a 100k volume pot for example, its my go to model for headphone amps and preamps.

Douglas Self was basically right though: at the end of the day you can't go far wrong with the 5532-5534. Just a really nice, easy to implement balance of low noise, low distortion, good drive, and stability.

As we all expected, if you calculate, the FET OP2134 should be the noisiest, followed indeed by the good 'old bipolar NE5532 and the very low noise LM4562.

However in practice, things get a bit different:
NE5532:
An externally hosted image should be here but it was not working when we last tested it.


OPA2134:
An externally hosted image should be here but it was not working when we last tested it.


LM4562:
An externally hosted image should be here but it was not working when we last tested it.


The NE5532 comes out best, followed closely by the OPA2134. The LM4562 is actually worse than both for this application. You can find the full measurements for all opamps here.

This is actually quite funny, because lots of people around here do not recommend the NE5532, because the impedance of an MM cart would be to high. It's not. It's a perfect choice. I do note however that the frequency response of the OPA2134 is "flatter". This causes a bit more bass to come through, and a bit less treble. This might explain why people call the OPA2134 a "warmer" sounding opamp.

NE5532 frequency response:
An externally hosted image should be here but it was not working when we last tested it.


OPA2134 frequency response:
An externally hosted image should be here but it was not working when we last tested it.
 

rjm

Member
Joined 2004
Paid Member
Interesting, but lets not get ahead of ourselves.

The predicted difference in the noise baseline is only about 3 dB between each successive op amp. That's difficult to see on an axis scaled to 140 dB and difficult to calculate given the many spurious peaks at 1-10 kHz.

Also, we must think about the PSRR - different op amps have very different values, one can be letting in some power supply noise where the other is not. That could reverse the measured result from the predicted trend.

Finally, a big one: GBWP. The LM4562 is a fast sucker, 55 Mhz. Too fast for this circuit really, it may likely have additional noise pickup from RFI interference and instability too.

Though I would add that from where I stand, all three would be acceptable and the choice would largely come down to preference.

I do not recommend using op amps with bandwidths over 10 MHz in any of my circuit boards, however. I have not tested for stability and the boards / power supply are not designed to operate at such high frequencies.
 
Last edited:

rjm

Member
Joined 2004
Paid Member
Except incorrect attempt adding two values ​​so far we have no evidence that the RIAA curve is correct..

vulejov, you can find the correct RIAA eq. curve here, together with the formula to calculate it. To get in the inverse, simply flip it by taking the negative of the dB value.

To anticipate your question "how can I know if it is correct or not?" let me answer that for you now:

Because I said so.

If you have further doubts, feel free to make your own investigations, ask the RIAA, whatever, I am not willing to waste any more of my time with this.
 
vulejov, you can find the correct RIAA eq. curve here, together with the formula to calculate it. To get in the inverse, simply flip it by taking the negative of the dB value.

To anticipate your question "how can I know if it is correct or not?" let me answer that for you now:

Because I said so.

If you have further doubts, feel free to make your own investigations, ask the RIAA, whatever, I am not willing to waste any more of my time with this.

It is easy to find the correct curve.. but it is not my point..

I thought about your answer based on any calculations that makes me sure that the RIAA curve in your preamp is correct, and that these are not simulations ..

Attempted addition of capacitance and resistance before a page is simply not correct..
 

rjm

Member
Joined 2004
Paid Member
impedance of a resistor Zr(f) = R
impedance of a capacitor Zc(f) = 1/(i 2 pi (f) C) [or "j" if you prefer]
impedance of a resistor and capacitor in series, Ztot(f) = Zr(f) + Zc(f)

So tell me, what is not correct about the above?

For everyone else reading: the beauty of impedances is that inductive and reactive elements in a network (inductors and capacitors, respectively) can be manipulated mathematically just like simple resistors, adding as Z1+Z2 in series and as Z1*Z2/(Z1+Z2) in parallel. The only difficulty is that the numbers themselves are complex, with real and imaginary parts. If you can use a computer to handle complex numbers, the analysis is straightforward.

Now I'm curious. I mean, I thought everyone who was into DIY knew this, even if they weren't inclined to work through it themselves.

Quick show of hands please: "the impedance of a capacitor is -j/([omega]C" .... "Yeah, I knew that" Y/N?
 
Last edited:
I'm not going to. It's not for me or anyone here to do that work for you, especially when LTSpice gives the result with no effort!! If you doubt it, it's your job to prove its wrong, not ours to show you its right!!

Here at least is the approach:

Divide the circuit into three parts: the input inverting op amp, output inverting op amp, and output RC filter. I will note the impedance of any capacitor as Z(C1) etc., defined as above.

First stage gain: easy, as it can be assumed flat over the audio band, so gain1=R3/(R1+Zin), Zin is the DC resistance of the cart.

Second stage, gain2= |[Z(C3)||R7||R8+Z(C2)]/R4| (two bars '||' means "in parallel with", single bar, e.g. |M| means (magnitude of M).

Third stage gain3= |(R9||Zout)/(Z(C1)+R9||Zout)| Zout is the impedance of the following stage, not shown on your schematic.

Total circuit gain in dB is 20 log (gain1*gain2*gain3).

The above leaves out some high frequency components but will be sufficient to allow you to calculate the gain up to 20 kHz with sufficient accuracy.

At least I think that's right. It's been a while and I do make mistakes.

Good luck!
 
Last edited:
Second stage, gain2= |[Z(C3)||R7||R8+Z(C2)]/R4| (two bars '||' means "in parallel with", single bar, e.g. |M| means (magnitude of M).

Third stage gain3= |(R9||Zout)/(Z(C1)+R9||Zout)| Zout is the impedance of the following stage, not shown on your schematic.

Total circuit gain in dB is 20 log (gain1*gain2*gain3).

The above leaves out some high frequency components but will be sufficient to allow you to calculate the gain up to 20 kHz with sufficient accuracy.

At least I think that's right. It's been a while and I do make mistakes.

Good luck!

What is Z in second stage?

Zout is 10k..