Active notch filter for THD+N measurements

[askii2]

Hi,

This is continuing with my earlier posting about measuring an amplifier's response. I plan to use it to measure the THD+N of an amplifier I'll build in a month's time. The measurement will be done by supplying a sine wave to the amplifier and measuring it's output minus the original wave. The notch filter posted here is for removing the sine wave. Since I won't be looking for a certification, I believe a notch filter with -50dB at the central frequency should suffice. Attached schematic shows my design. I'll very much appreciate feedback from
experts on this forum. Thanks in advance.

Input - Line level sine wave for validating the filter. Amplifier output after passing through a 1k potentiometer acting as a voltage devider.

Output - Connected to line in of a laptop for analysis using a software oscilloscope.

Section 1 - MAX7410 based notch filter. This was inspired by a design idea from maxim at this website Switched-Capacitor IC Forms Notch Filter - Maxim Frequency is decided by the capacitor connected to pin 8. It's set to 100Hz in this design. I could potentially add a switch here to get frequencies from 25Hz to 15kHz. Center frequency sees a gain of -50dB. higher frequencies are unchanged. There is an upto 3dB again for lower frequencies. Since THD is about harmonics, the gain at lower frequencies shouldn't matter IMHO.

Sections 2 and 3 - LM358 based amplifiers. Each of them has a gain of 50. They should be flat over the whole audio audio range.

Protection - 4.7v zener diodes at the input protect from high input voltages. Diodes near the output protect against reverse DC or AC connections. I couldn't build a short circuit protection here. Any ideas on a simple short circuit protection scheme?

[Frank Berry]

Interesting design.
I'm not sure that a -50dB notch is deep enough to get a meaningful reading of the harmonics.
You might want to re-think the low frequency gain as well. Power Supply noise is an important part of the THD+N measurements.
The typical Harmonic Distortion meter uses an adjustable twin-tee notch filter in a configuration where an Operational Amplifier provides a low impedance feedback path to the filter to maximize the depth of the null and to make the null more narrow.

Have fun with your project. By trimming some of the component values, you might be able to obtain a deeper null.

[askii2]

Thanks Frank.

Originally Posted by Frank Berry

Interesting design.
I'm not sure that a -50dB notch is deep enough to get a meaningful reading of the harmonics.

Most off the shelf good quality amplifiers have 0.01% or better THD+N figures over whole audio range at 80% or higher rated power. 0.01% is -40dB. So I believe -50dB notch should suffice. Is that correct?

On a related note - I of course by no means think my first amplifier will have as good a THD measurement, but I still need to have tools that'll show how good/bad my amp is .

Originally Posted by Frank Berry

You might want to re-think the low frequency gain as well.

I have been worried about that. That's why I have got first stage capacitor 4.7uF. It's reactance 1693 ohm at 20Hz which isn't too less compared to the 5kOhm resistance it connects to.

I have also gotten 0.1uF in each amplifier stage. It's reactance 80k ohm at 20Hz, which is comparable to the 100Hz resistance it's in series with.

The same figures aren't as bad at 100Hz (338Ohm and 16kOhm).

Will increasing these capacitances be good? Say 47uF for the first stage and 4.7uF for second and third stages. First capacitor needs to be high quality since it's in signal path. I wonder how much a 47uF tantalum or polypropylene cap costs.

Originally Posted by Frank Berry

Power Supply noise is an important part of the THD+N measurements.

Good point. I'll get a regulated 5V power supply chip on the board, then.

Originally Posted by Frank Berry

The typical Harmonic Distortion meter uses an adjustable twin-tee notch filter in a configuration where an Operational Amplifier provides a low impedance feedback path to the filter to maximize the depth of the null and to make the null more narrow.

I read several designs on the net. The twin-tee based on LM102 from National Semiconductor's Linear Brief 5, March 1969 requires closely matched components.

The bi-quad notch filter from LM148's data sheet features an adjustable Q, but getting it to work over the whole audio range is very difficult. Besides there are so many resistances and capacitances that finding component values also becomes a challenge.

Originally Posted by Frank Berry

Have fun with your project. By trimming some of the component values, you might be able to obtain a deeper null.

Not sure I understand that. Could you tell me more about how to obtain a deeper null?

[coluke]

Originally Posted by askii2

Most off the shelf good quality amplifiers have 0.01% or better THD+N figures over whole audio range at 80% or higher rated power. 0.01% is -40dB. So I believe -50dB notch should suffice. Is that correct?

Nope - 0.01% is -80dB, you need a 90dB deep - at least: the deeper, the better - * stable * notch in order to visualize distortion products.

Not sure I understand that. Could you tell me more about how to obtain a deeper null?

I'd suggest you to try a Bainter notch, or an high-pass SVF notch, running at moderate Q - both require 3 (or 4) opamps, but are fare less senstive to passive component matching than the bootstrapped Twin-T or Wien bridge notches. Finally, throw away those poor old LM358, and use instead some good low-noise/low-THD opamps (5532, 5534, OPA2134, LM4562...) - THD in the post-notch amplifier is not exactly a big issue in the 0.01% range, but the lower it is, the better (the same holds for noise, so lower the resistance in the feedback legs...).

Ciao,

L.

[Bigun]

I'm not sure how fancy or quantitative a measurement of distortion you really need and whether you'd settle for something that might not be definitive but could be a lot easier to do ?

"It may surprise you to know that one can quite easily perform cursory distortion analysis on this (and many) power amplifiers using an 8ohm power resistor, an oscilloscope, oscillator, and the amplifier itself."

from:

http://www.passdiy.com/pdf/citation.pdf

[jcx]

fft software and soundcards can easily resolve harmonics @ -80 dB without notching out the fundamental

simpler nulling of 20-30 dB will get you plenty of resolution

[coluke]

Originally Posted by jcx

fft software and soundcards can easily resolve harmonics @ -80 dB without notching out the fundamental

simpler nulling of 20-30 dB will get you plenty of resolution

Pretty sure, but I presume askii2 is managing to watch something like this:



Well, I hope he's not going to see exactly this, of course (x-over residuals from an LM358-based unity gain follower without pull-down resistor - SVF-based auto tuning/nulling notch, stable notch depth ~100dB, post-notch gain +40dB).

L.

[dchisholm]

Yes, if you're at all serious about making measurements then drop the LM358. The TL07x series outperform it across the board at about the same cost, and the NE5534 beats the pants off it while still falling in the "low cost" category.

And a 50 dB notch depth won't be adequate for evaluating the performance of any modern audio amp if the output of your notch filter simply goes to a broadband detector or meter. That barely gets you to an accurate measurement of 1% (THD + N). On the other hand - if this notch filter is used as a front-end to a spectrum analyzer or wavemeter that already has a resolution of -60 dBc (or much better), then you are starting to get into the range where your measurement capability is comparable to state-of-the-art performance.

I have no experience with that Maxim chip, but I can appreciate the desire to make high-resolution distortion measurements without a lot of carefully matched components and other attention to multiple details. My fear would be that the switched-capacitor filter will add corruption to the signal and limit your measurement floor far short of where you're trying to get.

For background information, you should probably take a look at Dick Moore's page, "About THD Analyzers", as well as his home-constructed notch filter at "Active Twin-T Notch Filter". The definitive article for DIY THD analyzer construction is Bob Cordell's 1981 series, "Build a High-Performance THD Analyzer (Part1, 2, and 3)", available on the Cordell Audio website at < CordellAudio.com - Home >.

Dale

[askii2]

I am not looking for publishing these statistics. It's for my own satisfaction that my amplifier is indeed good. I don't want to rely on my ears alone for this test.

I agree, it's a great way to test an amplifier. I think this should suffice for now. I don't need to build expensive or time consuming circuits as well . Thanks for this reference.

Originally Posted by Bigun

I'm not sure how fancy or quantitative a measurement of distortion you really need and whether you'd settle for something that might not be definitive but could be a lot easier to do ?

"It may surprise you to know that one can quite easily perform cursory distortion analysis on this (and many) power amplifiers using an 8ohm power resistor, an oscilloscope, oscillator, and the amplifier itself."

from:

http://www.passdiy.com/pdf/citation.pdf

[askii2]

Originally Posted by jcx

fft software and soundcards can easily resolve harmonics @ -80 dB without notching out the fundamental

That's what I had tried first. My earlier attempts (elliptical response curves I had posted earlier) were limited to transfer function measurements only. I measured a zero THD + N! Which is why I was trying a notch filter to get better resolution.

Originally Posted by jcx

simpler nulling of 20-30 dB will get you plenty of resolution

This may work, thanks.