Arb generator with tracking filter?

[precaud]

Does anyone make one? An AWG with anti-aliasing or decimation filter that tracks the sample playback rate?

The closest I've found is the old Lecroy 9100 series, which had selectable output filters.

[David Hess]

That is actually a pretty difficult task.  I have seen switched capacitor filters used as tracking antialiasing filters in waveform generation but they are only suitable up to 100s of kHz.  I think transconductance based variable filters could be used up to several MHz.  I suspect it is just easier to increase the sample rate relaxing restrictions on a fixed antialiasing filter.

[precaud]

No doubt difficult... I've never seen it. But I want one 

"Oversampling" for lower freqs works as long as you can stay within the AWG's sample length limits. Perhaps for now I just need to create versions of the stimulus (a "chirp sweep") at different sample rates/lengths and switch between them.

[pigrew]

Out of curiosity, what is the specification you are trying to meet with the filter?

I only know of the old Lecroy/Tabor generators which you already mentioned. SRS has a generator which also switches filters, but it does not do arbitrary waveforms.

Some generators allow the sample clock rate to be varied (Some Tabor generators, and also Siglent IIRC). This may better fit your needs.... Not sure.

[precaud]

No hard spec, just filling in the stairsteps between points, which wreak havoc on the FFT'd results when jitter from any cause is present. Renders ADC ENOB specs irrelevant!

The DACs I have on hand all have variable clock rates, and that indeed invites the problem. IIRC, they all have fixed output filters at or near 1/(max Fs/2). The stairstepping increases as the clock rate is slowed below max Fs.

[nctnico]

Why not turn the problem around? You know the samplerate so you can discard the FFT results which are at the samplerate or above.

[precaud]

Experience has shown that, for good results with this measurement, the sample playback rate has to be an integer multiple of the ADC clockrate.

[pigrew]

Have you tried the TrueForm generators from HPAK (33500B,33600A)? They perform digital interpolation and filtering, prior to the DAC. This should reduce the need for an analog filter.

[precaud]

Never heard of 'em, I'll look into it, thanks.

[nfmax]

I just scored a used Agilent 33522B and had a quick play with it. The TrueForm filter does exactly what you say you are after. With the filter enabled, the cutoff frequency tracks the programmed ARB sample rate. You get a choice of two filters, 'normal' with good flatness but some overshoot, and a -3dB point of 0.27, and a 'step' filter with a monotonic transient response and a -3dB cutoff of 0.13 (of the sample rate). With either filter, playing back an ARB on the scope gives a 'smooth' curve on the oscilloscope. Turning off the filter shows stair-steps as you would expect.

Maximum sample rate is 250 MS/s

[precaud]

I just scored a used Agilent 33522B and had a quick play with it. The TrueForm filter does exactly what you say you are after. With the filter enabled, the cutoff frequency tracks the programmed ARB sample rate.

I have looked online but have been unable to find anything that documents/describes the "tracking filter" behavior with arbs. Interestingly, Keysight doesn't even mention it in their promo videos. Can you point me to something that does?

[nfmax]

It's explained in the 33500 series Operating & Service manual, document 33500-90901, on page 178. Available from Keysight at https://www.keysight.com/main/redirector.jspx?action=ref&cname=EDITORIAL&ckey=2197440&lc=eng&cc=GB&nfr=-536902257.1026944.00

[precaud]

Great, thank you.

For anyone else interested, here is the filter description from the manual:

"Each filter’s cutoff frequency is a fixed fraction of the waveform’s sample rate. The Normal filter’s response is -3 dB at 27% of the sample rate and the Step filter’s response is -3 dB at 13% of the sample rate. For example, for an arbitrary waveform at 100 MSa/s, the Normal filter’s -3 dB frequency bandwidth is 27MHz."

That is excellent. Would love to see how they're doing this, over the entire bandwidth of the instrument.