8-bit samples on USRP2 at 33.333Msps

GNU Radio
1

Just did a quick test of 8-bit samples on a USRP2, on one of my 6-core
AMD machines.

I can get to 33Msps with a fairly simple SOURCE–>MULT–>FFT_SINK
flow-graph, but anything more complicated causes overruns.

If I try 50Msps, I get near-continuous overruns.

This is on an AMD Phenom II X6-1055 – the cores run at 2.8GHz, and I
have 4GB of 1333MHz memory.

It’s a fairly-wimpy machine as multi-core DSP monsters go, but I thought
I’d give it a try.

People should be aware that since this uses the upper-half (as far as I
know) of the ADC, you’ll need more gain to drive those
8 bits.

–
Marcus L.
Principal Investigator
Shirleys Bay Radio Astronomy Consortium

2

On 11/09/2011 07:07 PM, Marcus D. Leech wrote:

It’s a fairly-wimpy machine as multi-core DSP monsters go, but I thought
I’d give it a try.

My AMD Athlon™ II X4 620 Processor does jyst fine with 50Msps FFT
plotter. Just sayin :slight_smile:

People should be aware that since this uses the upper-half (as far as I
know) of the ADC, you’ll need more gain to drive those
8 bits.

There is a configurable scale factor (see the stream args in the grc
block). It defaults to a value that gives you the upper 8.

-Josh

3

There is a configurable scale factor (see the stream args in the grc
block). It defaults to a value that gives you the upper 8.

-Josh

I noticed there was a “scalar=1024” option. Is that just a multiplier
that UHD applies on the host
side for 8-bit samples, or does it affect the hardware (a barrel
shifter or something).

Its a full complex scaler multiplier in the FPGA. You can pick any
integer. Basically extract any of the bits you want.

-Josh

4

My AMD Athlon™ II X4 620 Processor does jyst fine with 50Msps FFT
plotter. Just sayin :slight_smile:

Hmmm, is the Athlon II generally better than the Phenom II (insns/clock?
). I suppose I should try it on my
1090T as well, which is rather faster than the 1055T I tried it on
tonight. But that means lifting the
1055T rack enclosure off of the 1090T, diving inside to pull the SD
card of the USRP2, etc, etc.
[My two USRP2 are mounted inside the 4U rack enclosures of their host
computers].
Perhaps tomorrow night :slight_smile:

People should be aware that since this uses the upper-half (as far as I
know) of the ADC, you’ll need more gain to drive those
8 bits.
There is a configurable scale factor (see the stream args in the grc
block). It defaults to a value that gives you the upper 8.

-Josh

I noticed there was a “scalar=1024” option. Is that just a multiplier
that UHD applies on the host
side for 8-bit samples, or does it affect the hardware (a barrel
shifter or something).

The point I was trying to make was that if your analog samples don’t
“light up” more than the bottom few bits, grabbing the
top bits (including sign) isn’t going to give you much useful
information in the samples (except the sign bit). A couple of years
ago when I was playing with 8-bit samples on the USRP1, I found I had
to increase the gain of my analog componentry to get
it to “light up” the top bits of the ADC, so that I had enough
information that was “in range”.

–
Marcus L.
Principal Investigator
Shirleys Bay Radio Astronomy Consortium