Some of you may have seen the "Blind Listening Tests & Amplifiers" thread in the Solid State Amplifiers forum. One of the things discussed there is null difference testing. Null testing is generally used one of the following ways:
Null testing can be used to compare the errors and distortions produced by two devices under test. You simply subtract the gain-matched ("nulled") output of the two devices from each other. If you get a deep null, a strong argument can be made the two devices will be indistinguishable in listening tests.
Null testing can also be used to compare the input to a device with the output of that device. If you subtract the gain-matched ("nulled") output from the input, what you're left with is ALL distortion and errors produced by that device. The resulting null signal can then be quantitatively measured in various ways (i.e. average level, spectral content, etc.). You can also listen to it if you want to subjectively evaluate the distortion products, etc.
It should be stressed that null testing allows the use of ANY kind of test signal ranging from real music to pink noise to sine waves. If you're testing an amplifier, you can test it driving real-world speakers.
Null testing allows the device under test to be used exactly as it would be when listening to music in your system. SO MOST OF THE USUAL ARGUMENTS AGAINST CONVENTIONAL MEASUREMENTS DON'T APPLY TO NULL TESTING. It measures ANY kind of distortion or errors produced by the device including frequency response deviation, phase shift, THD, IMD, TIM, etc.
A famous example using null testing was the Carver Challenge in Stereophile. Bob Carver challenged the editors of Stereophile they would not be able to hear any difference between a $700 Carver amp and whatever amp they wanted to provide as a reference amp. Stereophile chose a very well regarded high-end tube amp with a five figure price tag. They were very certain they would win the challenge.
Bob Carver spent a few days nulling his $700 Carver amp against the reference amp. He added some series resistance to the output and tweaked the feedback loop in the Carver. He claims he ultimately achieved a -70db null between the two amplifiers. The editors at Stereophile listened for two days and finally gave up--they couldn't hear any difference!
The Carver Challenge, IMHO, further validates null testing as a very powerful tool for audio design. The technique was first published by Baxandall in 1977 and then again in 1986 by David Hafler. You can find diagrams of the null setups proposed by both authors in this link:
http://www.dself.demon.co.uk/subjectv.htm
I've attached a GIF of a simple null setup that most DIY folks can easily try. Ideally, the 50 ohm pots should be multi-turn (trimpots are fine) so you can get an exact null but single-turn pots can also work.
A test CD with a pink noise track makes an ideal signal source. Pink noise is a dynamic signal that contains all frequencies in the audio band in roughly the same proportions as real music. You can, of course, also use real music and sine wave test tracks.
You simply hook a sensitive AC voltmeter across the null terminals, set a reference level at one of the reference points and then adjust the pots for the lowest voltage at the null terminals.
I put together the setup shown in the diagram and got the following null results:
Reference Level: 10 volts rms (after the divider network)
Null Level 100hz sine wave: 1.8mv (-75db)
Null Level 1khz sine wave: 2.0mv (-74db)
Null Level 20khz sine wave: 9.1mv (-61db)
Null Level 100hz square wave: 3.5mv (-69db)
Null Level Pink Noise: 9.8mv (-60db)
Null Level Music (peak hold): 8.1mv (-62db)
The above measurements compare the left and right channels of the preamp and power amp signal chains. Any deviation between them is at least 60db below the audio signal. As you can see, the channel match is over 10db worse at high frequencies--probably due to component tolerances in the feedback loop and board layout issues in the preamp and amp. NULL TESTING IS VERY SENSITIVE!
Now, just for fun, I swapped out ONE of the interconnects between the pre-amp and power amp. I used a cheap $2 RCA thin patch cord in one channel and a $140 KimberKable "Hero" in the other channel. I repeated the null measurements and they were all within 1db of the ones listed above.
When I have more time, I may try to perform a spectral analysis of the null signals with both cables. But just looking at the depth of the null, I'd have to say the cheap interconnect and the expensive Kimber interconnect are essentially equal and hence I doubt anyone would be able to hear a difference in a proper blind listening test.
In addition to testing cables, this setup can also be used to compare changes to an amp or preamp. You can change capacitors, resistors, etc., in one channel and compare the null results. If there's no significant change, it's a safe bet the change you made didn't make an audible difference.
Some things to be careful of... You ideally need a meter that's reasonably flat out to 20khz for the most accurate measurements. Many cheap DMMs are not. If you want to view the signal on a scope or perform a spectral analysis, be careful of grounding issues. The 1k resistors should prevent any amplifier damage if you ground one of the outputs, but you will get invalid measurements. Be careful of ground loops, etc.
Null Difference Testing
