Tone Bursts as Test Signals

Frank Koenig

Sophomore
Mar 7, 2011
187
0
16
Palo Alto, CA USA
www.dunmovin.com
Time-frequency analysis lets us look at the same thing many different ways, and I've always been intrigued by getting to the same place by different paths, both literally, say when walking or driving, and metaphorically. Tone bursts seem like natural test signals as they resemble the musical waveforms we're interested in reproducing, and I've toyed with them in various ways in the past to represent linear system behavior. I have here a really simple representation that might serve as a "common sense" check for crossover networks, if not other things. It's like a time-frequency-energy representation with teeth (the oscillations of the tone burst itself).

I take the impulse response of the system under consideration, which usually consists of an acoustic impulse response measurement processed with simulations of the proposed filters, and convolve it with a set of tone bursts of logarithmically spaced center frequencies and a fixed number of cycles. I then display the resulting time responses as a set of adjacent columns where color represents amplitude and each column corresponds to a particular frequency. It's as if we're viewing a set of adjacent oscilloscope traces from the top.

Below are 3 examples of the same crossover properly aligned, off by 1/2 cycle, and off by one full cycle at the crossover frequency of ~435 Hz. While in this case the correct alignment is pretty obvious from the phase traces, which are well behaved, this might be a useful check on messier (sub) alignments where we want to be sure we're not off by an entire cycle, or could do better by flipping polarity.

I'm using a raised cosine (von Hann) window to form the tone bursts although I doubt it's critical. Since a tone burst is the product of a sine wave and a window its spectrum is the convolution of an impulse (at the positive and negative sine wave frequencies) and the Fourier transform of the window, so just the shifted spectrum of the window. Since the spectrum is narrow and the window's sidebands are way down, the tone burst tends to suppress aliasing in the frequency domain. This is of little consequence when the measured signal is band limited to begin with, as it usually is, but is a useful property when looking at the tone burst response of high-pass systems alone, as when looking at how a particular filter behaves with an ideal speaker.

As an aside, a few weeks ago I spent some time feeding (rectangular) tone bursts from signal generator into a sound system and looking at the response with a mic connected to a scope. It's an instructive thing to do although the tick-tick-tick, dunk-dunk-dunk will drive you nuts after a while. Try it sometime.

Best,

--Frank96TBgood.jpg96TBhalfCycle.jpg96TB1cycle.jpg
 
Re: Tone Bursts as Test Signals

It's not very intuitive but yes the on/off step function is significant... One of my early technician jobs back in the '70s dealt with speech that was sampled, pitch shifted (stretched out, or squeezed tighter), then re-assembled together either with empty gaps or too much signal that gets discarded. Splicing these samples together without sounding like crap is not trivial.

So not exactly sine waves (tone bursts), but the perturbations from winding down the one sample and winding up the next was very revealing about the sound of the on/off step function, effectively multiplying the audio signal present by 1x or 0x. Years later I did a lot of work with tone bursts while designing dynamics circuits, like compressors, noise gates, companding NR, etc. A tone burst is an extremely difficult and revealing stimulus signal for gain changing circuitry.

One of the things I learned pretty early on when trying to use tone bursts to parse designs, was that how it looked on a scope display, rarely correlated with how it sounded. While I generally defer to bench testing over listening tests, tone bursts are my one exception, when listening trumps looking. To use tone bursts for speaker (listening) testing, first we must come up with a reference benchmark. So perhaps something like electrostatic headphones (without crossovers) could train you what a perfect tone burst should sound like. Then you put the same signal through a loudspeaker and listen for differences, either extra artifacts, or missing stuff.

The optimal tone burst for testing a given speaker will involve numerous variants in and bracketing the crossover region. Back in the day I designed my own tone burst equipment because available burst generators were limited in options. My tone burst rig consisted of two parts, a standard sine wave generator, and a burst gating device (your windowing). From my experience with splicing audio samples, I designed my gate to open and close at zero crossings (the step function of 1x times the 0V at zero crossings, eliminates clicks). Next I counted zero crossings so I only gated full cycles on or off (this eliminates inadvertent DC content from a half cycle). Finally my last feature was a dry (ungated) mix, so I could make the burst gentler than a full step on from complete silence by varying the ratio of gated to dry, more like real world signals.

Since my gate box was agnostic about what signal I put through it, I could use pre-recorded music as a source and vary the ratio between gated and dry to punch up the crest factor of normal music to be far more extreme than normal. If i can get my dynamics processor to handle this hyped up crest factor music, normal music is easy. The hard thing about meaningful listening tests, is getting difficult enough test signals to reveal flaws while still being realistic, real world. (Raw tone bursts do not occur in the real world. Nothing starts or stops instantly).

So to repeat, consider actually listening to your bursts playing through the speakers, but first learn what the actual bursts "should" sound like with good headphones.

JR

PS: Note this date on your calendars. This is the first and last time I will ever advise a listening test over bench testing.
 
Re: Tone Bursts as Test Signals

John, thanks. Interesting stories. I'm using an integer number of cycles and align the endpoints of the window with the zero crossings of the sine wave. I tried rectangular and Hamming windows and like the performance of the Hamming best right now. 7 cycles appears a good compromise between time and frequency resolution. I played a little with the color palette for better contrast. Black is zero with negative values represented by cyan, blue, and magenta and positive values by green, yellow and red. Examples below for aligned, LF off 1/2 cycle, and LF off 1 cycle.

Best,

--Frank
XO405good.jpgXO405halfCycle-Early.jpgXO405oneCycleEarly.jpg