How to measure frequence response...

[Neeno]

I've just finished my ssl9k preamps, i've done some test with my mics... they sounds better than compared to my presouns digimax... a particular sound...

Now i want to do some measures to compare them to my other preamps... but i really don't know where to start...

I want to measure the frequency response, THD, etc...

I have a waveform generator, an oscilloscope and some meters...

Anyone can help me ?

 

[mcs]

If you have an RMS voltmeter with a dB scale, measuring the frequency response is very easy. Set your oscillator to 1kHz and set the level so the meter shows 0dB. Now turn the frequency knob up and down to find the -3dB points. Be sure it's not your source you're measuring. Many preamps will go up to 2-500kHz without any problems...

You can find a simple distortion "instrument" here: http://members.aol.com/sbench102/TubeMisc/notch.gif

Best regards,

Mikkel C. Simonsen

 

[SSLtech]

I'd be very surprised if the SSL 9k preamp shows any drop-off short of radio frequencies at the top end...

The low end is the only place that you might be able to measure any rolloff, and even then only if your test gear goes that low...

The expected flat response of most transformerless mic pres that I see nowadays is beyond 100kHz... There's going to be more to the sound than just frequency response...

Keith

 

[NewYorkDave]

Important to note that the device under test should be driven from the correct source impedance, have the correct load impedance connected to its output, and tests should be done at the appropriate level. For instance, you usually wouldn't test a mic preamp with a +4dBU input signal from a 10K source impedance :wink: You would use (typically) a level of about -40dBU from a 150 or 200-ohm source impedance.

 

[NewYorkDave]

[quote author="kubi"]You schould also do a scan without the pre, just the cart's output connected to its input in order to see, what your cart adds to the SSL's figures and to be able to calculate the figures for just the SSL alone.[/quote]

Subtracting test instrument residual noise and distortion from the final measured result doesn't really work, since noise and distortion does not add in a simple linear fashion. The proper way to do it is to use instruments with very low residuals and do not attempt to "correct" the measured figures.

I do not think the typical 16-bit SoundMasher card found in most PCs is a suitable test instrument for quality audio equipment.

 

[slash14]

What's the best way to achieve frequency measurement ?

- injecting a dirac signal into the device to test and then compute the FFT of the output signal ?
- injecting some white noise, then compute the FFT of the ouptut signal ?
- injecting a frequency ramp and recording the output amplitude and phase ?

I know that injecting a particular signal into a device wil give you its impulsive response, and then with some mathematics, you can get the frequency response of the device from its impulsive response. Or maybe I am wrong and haven't understood all treatment courses I've been fed with during school

 

[NewYorkDave]

1. What's a "dirac" signal?

2. This is often done, but with pink noise (equal power-per-octave) instead of white noise (equal-power-per-bandwidth).

3. A frequency sweep and amplitude measurement is the most common method. Frequency response is usually given as relative amplitude figure without regard to phase.

 

[slash14]

[quote author="NewYorkDave"]1. What's a "dirac" signal?[/quote]
Oh, sorry, seems that "Dirac signal" is not called like this in english
A "Dirac signal" is a rectangular signal but with a very small duration. Something like that:

[quote author="NewYorkDave"]2. This is often done, but with pink noise (equal power-per-octave) instead of white noise (equal-power-per-bandwidth).

Frequency response is almost always plotted log-log.[/quote]What's the reason using equal power-per-octave instead of equal-power-per-bandwidth ? After all, octave is a particular case of bandwidth, isn't it ?

[quote author="NewYorkDave"]3. A frequency sweep and amplitude measurement is the most common method. Frequency response is usually given as relative amplitude figure without regard to phase.[/quote]
Is there a particular reason using square waveform signal tests instead of sine waveform ?

Last question: in many speaker test benchs, a capacitor is added to the speaker (bypassing the speaker to ground, isn't it ?). What's the aim of this operation ? Getting resonnance between the resultant RLC circuit (loudspeaker resistance + inductance coil + capacitor) ?

 

[NewYorkDave]

[quote author="slash14"]What's the reason using equal power-per-octave instead of equal-power-per-bandwidth ? [/quote]

Because white noise has equal power per Hertz, so the power doubles for each octave, and a device with a flat frequency response would NOT give a straight line if plotted with white noise.

Is there a particular reason using square waveform signal tests instead of sine waveform ?

I didn't say anything about using a square wave. I test frequency response using a swept sine. A square wave can be used to spot some gross anomalies in frequency response, though, by looking at the shape/tilt of the waveform.

Last question: in many speaker test benchs, a capacitor is added to the speaker (bypassing the speaker to ground, isn't it ?). What's the aim of this operation ? Getting resonnance between the resultant RLC circuit (loudspeaker resistance + inductance coil + capacitor) ?

I've never worked in a speaker lab, so your guess is as good as mine. I guess adding a cap to lower the self-resonant frequency of the speaker must serve some kind of purpose in these tests. Can you post a link to an example?

 

[slash14]

[quote author="NewYorkDave"]Because white noise has equal power per Hertz, so the power doubles for each octave, and a device with a flat frequency response would NOT give a straight line if plotted with white noise.[/quote]
OK ....



[quote author="NewYorkDave"]I didn't say anything about using a square wave. [/quote]
Yes, I know you didn't talked about sqare waveforms in your previous post but the question came to me while I was writing mine
I'll keep on dealing with square wave: quiet often, square waves are used to measure time rise and fall time (I don't know if it's the right terms), isn'it ? Can't we guess the bandwidth from this rise time ?

Last question concerning square waves testbench: most of the test benchs I read about hifi and homecinema devices were done with square waves and not sine wave. In my opinion, square waves are better to test bandwidth as they have some horizontal stages which is interesting for testing devices at very low frequency (as DC is 0Hz, isn' it ?). And finally, square waves have some very small rise time and fall time which imply very high frquencies ...
Sine wave should be used to test THD.

Well, it's just an idea coming to me

I hope I've been understood, if not, tel me

 

[NewYorkDave]

Looking at a square wave will give you a qualitative idea of frequency response to an extent--particularly high frequency response--but it's not quantitative. For a meaningful measurement, you need to be able to say that the amplitude response is flat +/- so many decibels from one frequency limit to another.

Don't forget, the square wave contains no content below its fundamental frequency; a 1kHz square won't tell you **** about your frequency response at 100Hz. And since no system has an infinite bandwidth, and a square wave is composed of an infinite number of odd harmonics as well as its fundamental frequency, the square wave will NEVER be reproduced exactly. If the frequency of the square wave is low enough, and system bandwidth wide enough, it can look pretty good on a scope.

Square waves are good for checking risetime and phase shift when applied properly. But for a plot of frequency response, pink noise and FFT, or swept sine and amplitude measurement, are the way to go.

 

[SSLtech]

A cap switched across a speaker seems a reasonable way to assess how excessive cable capacitance might start to affect measurements when the amplifier isn't six inches away from the speaker... as are resistances and inductances in series...

Just a guess...

Keith

 

[PRR]

Dave's notes about source and load impedance are VERY important. Especially with transformers, but other devices can be very critical of impedance.

> in many speaker test benchs, a capacitor is added to the speaker

There are (or used to be) amplifiers that worked well on a resistor, or a small inductor in series with a resistor, but got very sick if they saw a capacitor. Speaker leads have some capacitance, some speaker crossovers act as capacitors, and there used to be a fad for Electrostats which is when everybody started checking amps with a capacitor.

> most of the test benchs I read about hifi and homecinema devices were done with square waves and not sine wave.

Sweep tests are always done with sines.

In pictures, square-waves make pretty shapes, so the illustration is often of a squarewave.

In theory, the complete frequency response can be derived from the square-wave output, if there is enough detail (oscilloscope traces are not all that sharp). In practice, it is a lot easier to swing the dial on the sine-wave generator and watch the meter, than to reverse-compute the frequency response from the square-wave response.

> the square wave contains no content below its fundamental frequency; a 1kHz square won't tell you **** about your frequency response at 100Hz.

It will. If the top "flat" part of a 1KHz square is tilted 10%, the frequency response is drooping by 100Hz. If tilted 1%, then I think about 10Hz. If you can prove that there is NO tilt, then you have shown that response extends "infinitely" below 1KHz.

 

[CJ]

One problem with square wave testing is finding a generator that will keep it's shape no matter what it is driving, and no matter what frequency. At low frequencies, the output of the generator is influenced by the load it is hooked to. At least with the generator I am using. I think this is why Jensen uses a synthesized waveform to test low end phase shift and response..

 

[slash14]

OK, thanks for all your answers.

If someone finds some old courses of signal treatment applied to music, I'd like to take a look at them please

I'm not affraid by mathematics .. well not too much

 

[NewYorkDave]

[quote author="PRR"]If the top "flat" part of a 1KHz square is tilted 10%, the frequency response is drooping by 100Hz. If tilted 1%, then I think about 10Hz. If you can prove that there is NO tilt, then you have shown that response extends "infinitely" below 1KHz.[/quote]

Actually, tilt will show you phase nonlinearity. Tilt on the leading edge means low frequencies are leading; on the falling edge means they're lagging. Rounding or scooping of the wave will show rising or falling LF response.

At any rate, it's kind of a pain in the ass to make quantitative frequency response measurements with square waves, compared to the ease of using a swept sine and a VTVM, or injecting pink noise and looking at a spectrum analyzer. Glancing at a square wave can tell you a lot, but measuring "how much" often involves a lot of timebase magnification and counting of tiny graticules :wink: