What could be causing this square wave distortion (pics inside)

[percy]

I have a Turtle Beach Santa Cruz soundcard and it gives pretty good results with RMAA. Recently I did a loop record (Line Out to Line In) on the soundcard using a short loop cable for a square wave signal. I used Speaker Workshop and this is result I get.

100Hz -

[percy]

1000 hz -

Question is - What are the possible causes of the square wave shape skewing or ringing at the edges. Its without load. I also tried with a small load (100-200ohm resistor) and it did not mess up the shape/ringing any further. So what could it be ? Bad psu decoupling ? or psu noise ?

Sine waves record just fine - i.e. I don't see any obvious distortions in the curve. I have used this software and soundcard for speaker measurements for a while now and it has been performing just great.

[jahonen]

What you see is a Gibbs phenomenon (truncating the squarewave fourier series results ringing). Square wave does not converge nicely on reconstruction filter at soundcard output, as the real squarewave will require infinite bandwidth to be perfect. So that looks perfectly normal to me.

Regards,
Janne

[percy]

hmm..reconstruction filter at the soundcard output.
Are your referring to the DAC's reconstruction filter or the output decoupling cap/filter after the output opamp ?

[jahonen]

Originally posted by percy
hmm..reconstruction filter at the soundcard output.
Are your referring to the DAC's reconstruction filter or the output decoupling cap/filter after the output opamp ?

I mean the DAC reconstruction oversampling filter (which is actually digital).

Regards,
Janne

[percy]

What would be the typical effective bandwidth in that case ?

Even if the upper limit of the bandwidth was like 20Khz then are you saying a 100hz square would look about the same if measured in a fully analog environment too (no filters, conversion and such) ?

What would typically a 100hz square wave look like when fed and measured through an amplifier of that amount of bandwidth ?

[Eva]

As it has been pointed out, there is nothing wrong with these waveforms, they are the expected analog output from a bandwidth limited digital system.

On the other hand, bandwidth limiting in a plain analog amplifier will manifest itself as transient rounding in the square wave.

Also, tilting of the flat tops of the wave in the 100Hz measurement is a sign of low frequency rolloff.

[percy]

Originally posted by Eva
On the other hand, bandwidth limiting in a plain analog amplifier will manifest itself as transient rounding in the square wave.

probably that is what confused me - because this looks more like "ringing" than "rounding".

Also, tilting of the flat tops of the wave in the 100Hz measurement is a sign of low frequency rolloff.

Interesting, the low frequency is rolloff is like -1db at 20hz (or even better), which was measured by rmaa.

[jahonen]

The ringing is due to very steep roll-off. Any faster than gaussian roll-off yields to ringing. Symmetric ringing implies linear phase, which is normal. Analog roll-off is generally quite slow in contrast, thus no ringing, just "rounding".

Practically usable bandwidth is just below nyquist frequency, up to 0.45*fs/2, or something close. It is important to note that sampling theory considers only stationary signals, transient signals must be considered separately. So closer you get to nyquist limit, slower the transient response becomes, relatively speaking.

Regards,
Janne

[AndrewT]

Hi,
the slope on the lower freqency (100Hz) square wave is an indicator of

low frequency rolloff

or is it?

Is it the high pass roll off that causes the slope?
Is it the phase difference in the low end that causes the slope?
Is it the phase difference resulting from the high pass that causes the slope?

I was told a long time ago that a digital oscilloscope needed a sampling frequency about 10 times the required analogue bandwidth. Surely when working in reverse a similar ratio should apply. 44 kilosamples/second (ks/s) limits the bandwidth to about 4 to 5kHz.

However since square waves are the odd harmonics going to infinity then errors increase as the square frequency approaches 1/9 or 1/11 of the usable bandwidth. This would imply a maximum (upper) limit for a digitally generated square wave using 44ks/s of about 400Hz.
Why then, do we see similar errors at 100Hz and 1000Hz (ignoring the slope/tilt) either side of my implied frequency limit? Could it be due to a bad algorithm?