Re: [Discuss-gnuradio] USB2 <-> fast ADC & DAC

[Joseph DiVerdi]

Dear Andy,

I have been working on using GNU-radio for some fairly specific scientific 
applications and although I cannot speak to the "completely generic" uses can 
offer some thoughts.

IMHO, offering two channels of simultaneously captured input and of 
simultaneously updated output are rather important. This presages your 
questions about the utility of baseband-only device. I and others have built 
downconvertors and upconvertors which bring us from RF to baseband and vice 
versa. While the use of such devices is not "completely generic" it comes very 
close. To support them it is most useful to provide and accept complex data 
(both real and imaginary channels) which are sampled at the same time (rather 
than interleaved) and emitted similarly.

I suspect (and hope!) that others will write on the sub-sampling aspects of 
data acquisition.

<snip>
>This is intended to be a completely generic USB-powered baseband 
>module with all processing done back at the PC in realtime.  Milosch 
>did some tests and found it was possible to transfer 20MBytes/sec 
>like this with latencies of under 2mS over USB2.  It is this transfer 
>rate which is the limiting factor, not the conversion rates on the 
>ADC or DACs.
<snip>

My particular application favors 200kSample/second (complex!) continuously 
(although I'm currently getting by with a dual channel, 24kSample/second sound 
card). This is by no means representative of the rest of the world and I 
suspect that you will learn from the group feedback that there are several 
natural popularly required bandwidths (and hence sampling rates) which will 
guide the design.

<snip>
> - I assumed that a baseband-only device like this is in fact useful 
>for this project.  However I see existing designs are interested in 
>outputting modulated RF at ~30MHz via an FPGA or DSP.  Is a 
>baseband-only device even useful?
<snip>

As I wrote above: absolutely! Although you may receive feedback that certain 
other direct RF devices are also highly desired.

<snip>
> - What resolution and speed ADC is needed to be considered useful?  
>We can transfer up to ~10Msps at 16 bits simplex, or ~5Msps at 16 
>bits duplex (ie, simultaneous tx and rx).  The choice of ADC and DAC 
>specs has obvious impact on the cost.
<snip>

Again, can only speak to my own application. I am working with 
100kSamples/second (complex) with no data output.

<snip>
> - What kind of analogue IO is most useful?  1Vp-p 50ohm input and 
>output?  Transformer coupled?  Filtered?
<snip>

Based on experience I always favor differential signal input and output, 
especially when operating at baseband. The quality of the data is more often 
assured over single-sided signals. There exist plenty of modestly priced, 
instrumentation amplifiers which would provide a high impedance front-end to 
the input. Providing space for a soldered or push-pin terminating resistor will 
afford your users with the option of their favored value.

Additionally, the differential connections obviate the need for transformers 
and offer down-to-DC input and output.

Similarly on the output, differential drivers with low impedance in a single IC 
package are readily available and IMHO are the way to go.

Considering "completely generic" would suggest no filtering so that your user 
would provide it. Adding switchable filters to your device would push up the 
cost of goods. :(

<snip>
> - I saw on the list AGC mentioned, is this desirable on this kind of 
>module to have programmable attenuation or amplification?
<snip>

Not needed for me - I would prefer to set the levels in my downconvertor.

Hope you find all this helpful.

Best regards,
Joseph
-- 
Joseph A. DiVerdi, Ph.D., M.B.A.          
http://xtrsystems.com/           970.980.5868 (voice) 
PGP Key ID: 0xD50A9E33