Why do I have a sampling rate limit of 33.333 Msps?

GNU Radio
1

Hello,

I expect my N210 to have a sampling rate up to 100 Msps. Instead I do
not
seem to go above 33.3333 MHz. Why? I don’t see why the LFRX board should
be
a limitation.

Thanks,

LD Zhang

2

The N210 samples at a fixed rate of 100Msps. However, the gigabit
Ethernet
transport limits the instantaneous bandwidth over the transport to
25Msps
in 16 bit sampling mode, or 50Msps in 8 bit sampling mode.

–n

3

Is there a way to get around this limit? I suppose you have to have
on-board memory to hold data temporarily. Unfortunately the N210 isn’t
like
the E100 which has the memory?

LD

4

I would like to explore the system performance from 10 to 100 Msps at an
interval of 10 MHz. I know that 20 Msps is probably sufficient. But it
would be good to survey the performance at different rates.

Since my interest is not the entire band but multiple signal at discrete
frequencies, one possibility is that one sample at say up to 100 MHz
internal to the USRP, but then pass the signal through a set of filter
banks, and then digital down convert so that signals can be much lower
frequencies. Then decimation is truly an option without losing SNR. The
end
product is a lower rate data set which is OK for Ethernet transport. Is
this a good approach? Is there anything wrong in this thinking? Or is
this
too difficult for the USRP? Is there a better approach? Suggestions are
appreciated.

LD

5

How fast do you need to sample? How many samples do you need?

–n

6

On Tue, Oct 2, 2012 at 4:25 PM, LD Zhang [email protected] wrote:

this a good approach? Is there anything wrong in this thinking? Or is this
too difficult for the USRP? Is there a better approach? Suggestions are
appreciated.

That’s a great idea. It’s also what the N210 is doing internally. The ADC
samples at a fixed 100Msps, and a digital downconverter followed by a
decimator (including filtering) selects a portion of that spectrum for
transport over the Ethernet bus. When you ask the N210 for a sample
rate,
it decimates and filters appropriately for that sample rate. When using
LFRX, the frequency you pick in UHD is the digital downconverter
frequency.

N210 has two independent DSP chains, so if you like, you can get two
parallel streams from LFRX, operating at different sample rates and
different DDC frequencies.

–n

7

Hi Nick,

Thanks for your helpful comments. I have followed up on this train of
thought on specifying a frequency for the USRP. I guess the advantage is
one can achieve the same level of over-sampling with lower sampling rate
with this technique when one shifts closer to the frequency of interest.
I
gave it a try in the USRP and have the following observations:

  1. The spectrum center essentially shifts to the specified frequency. No
    band pass filtering seems to be applied at all. My only concern is that
    this lack of band pass filtering may hinder dynamic range in certain
    cases
    but not a problem for now.

  2. I have kind of a logistic question. I am running the GRC file which
    directs the output to a file sink. Since I am doing different parameter
    studies of the system performance, I need to generate different files
    with
    different operating parameters. I don’t know how to automatically dump
    parameters like samp rate, samp duration, center freq, BW, gain, etc.,
    to
    the recorded file so that analysis can be somewhat automated. Maybe this
    is
    a question that should be asked under a different subject heading.

Thanks,

LD

8

On Wed, Oct 3, 2012 at 2:00 AM, LD Zhang [email protected] wrote:

lack of band pass filtering may hinder dynamic range in certain cases but
not a problem for now.

Yes, that’s right. The daughterboards are designed to have a wide
range in frequency. If you are concerned about this issue, the idea is
that you will have to provide your own filtering (and probably
amplification) for a particular frequency band you are interested in.

  1. I have kind of a logistic question. I am running the GRC file which
    directs the output to a file sink. Since I am doing different parameter
    studies of the system performance, I need to generate different files with
    different operating parameters. I don’t know how to automatically dump
    parameters like samp rate, samp duration, center freq, BW, gain, etc., to
    the recorded file so that analysis can be somewhat automated. Maybe this is
    a question that should be asked under a different subject heading.

There is nothing that really does that right now. It’s been on the
to-do list for a while, but the issue is doing it right. There’s a lot
of history about how to store metadata in sample files, and we haven’t
wanted to lock ourselves into a way of doing this without using
something that people would find the most useful.

Tom