On 19/02/2025 02:22, Brendan Horsfield wrote:
Just to clarify one point: How do you define the start of the transition region? Do you go from the 3 dB corner frequency, or something else, like the equiripple bandwidth of the FIR filter?
I just did it visually on an FFT display from the 3dB corner.
On Wed, 19 Feb 2025 at 13:11, Marcus D. Leech <[email protected]> wrote:On 18/02/2025 21:45, Brendan Horsfield wrote:Point taken. At this stage we are mainly interested in straight IQ recording & playback with minimal processing. However, in the future it would be desirable to be able to display a real-time spectrum trace & waterfall plot during recording/playback, using GNU Radio or something like it. As you suggest, I am assuming our host machine will need a dual-10Gbe adaptor card and a high-spec CPU, memory, SSD etc. This is a complex procurement exercise all by itself.My understanding (and I haven't played with them) is that NVME SSDs are among the fastest. Performance up to a few GByte/Sec on write is possible, although I don't know if it can be sustained at those rates, or whether it's "bursty". I've been able to produce "real-time" spectral displays on 10yo dual-Xeon hardware at 100Msps, but only using the kind of "stuttered" display approach that Gnu Radio FFT displays often use, where most of the data is discarded. Often, that's all that's needed to show a quick summary of the spectrum. On your other question, about transition bandwidth, I don't have a direct answer, but on an N310 I measured the roll-off as a fraction of the overall bandwidth, and it is about 12.5%. That doesn't necessarily translate to the X310, but the DDC implementation is of the same generation.On Wed, 19 Feb 2025 at 10:58, Marcus D. Leech <[email protected]> wrote: On 18/02/2025 19:26, Brendan Horsfield wrote:I thought your name sounded familiar! 🙂 Overall the X310+UBX-160 appears to be a good fit to our requirements. My original question was really about ensuring that our host PC & network interface have sufficient bandwidth to ingest the IQ data from a pair of UBX-160s. It would be nice (although not essential) if we could run one channel at 100 Msps, and the other at 200 Msps, to reduce the bandwidth requirements on the backend hardware.You'd need to have separate streamers to support two different sample rates, and two 10Gbe interfaces. But in terms of "what kind of PC hardware do I need?". There's no closed-form answer to that question. There's no handy-dandy "engineering worksheet" that tells you how much "grunt" you need for different DSP "flows" at a given sample-rate--so very much depends on what you're doing, and how you're doing it. Generally, as you scale up in sample-rate, you have to scale up in: o CPU base clock rate o Memory bandwidth o Number of CPUsOn Wed, 19 Feb 2025 at 10:17, Marcus D. Leech <[email protected]> wrote: On 18/02/2025 19:13, Brendan Horsfield wrote:Thanks for the suggestion about the noise source -- that's what I would normally do. Unfortunately I haven't actually purchased the hardware yet -- I was hoping to clarify this issue before raising a purchase order. Perhaps I should follow this up with one of the application engineers at NI? They might have access to an X310+UBX-160 system that they can use to answer my question directly. Thanks again for your help in this matter. Regards, Brendan.I actually do work for NI on USRP devices (on a very very very part-time basis). My X310 is currently elsewhere, and not populated with a UBX-160.On Wed, 19 Feb 2025 at 09:55, Marcus D. Leech <[email protected]> wrote: On 18/02/2025 18:45, Brendan Horsfield wrote:Yes, I assumed that was the case. However, it is not clear from the X300 documentation how sharp those filters are. Can you tell me how wide the transition band is at the lower sample rates? To give you some context, I would like to use an X300 (or X310) with a UBX-160 daughterboard to digitise the entire 2.4 GHz Wi-Fi band, which is 83.5 MHz wide. Ideally I would like to use a sample rate of 100 Msps to minimise the data rate between the USRP and the host PC. However, before I do this I need to be certain that the usable bandwidth at this sample rate will be greater than 83.5 MHz. Is this information documented somewhere?It somewhat depends on the decimation. If the decimation has a factor of two or 4, the edge roll-off is fairly sharp. Otherwise, there's a half-band filter in-place that causes a less-desirable pass-band. But I don't know, precisely, what the transition band is in the "nicer" filter shapes. If you have an X310+UBX-160, you can always just use a noise source, and measure it yourself to see if it's appropriate for your application.On Tue, 18 Feb 2025 at 23:11, Marcus D Leech <[email protected]> wrote: There will always be some edge roll off. Decimation includes filtering and those filters cannot be infinitely steep. Sent from my iPhone > On Feb 18, 2025, at 2:12 AM, Brendan Horsfield <[email protected]> wrote: > > > Hi All, > > I have a question about the usable bandwidth of the X300 USRP / UBX-160 daughterboard combo at sampling rates below 200 Msps: > > As I understand it, the UBX-160 receiver has an analog (hardware) filter before the ADC that limits the usable bandwidth to 160 MHz, while the ADC runs at 200 Msps. Therefore the usable bandwidth is around 80% of the sample rate. > > My question is: What is the usable bandwidth at lower sampling rates? Does the 80% factor always apply? > > For example, if I set the decimation factor to 4, so that my sampling rate is 50 Msps, does this mean that the usable bandwidth will be 40 MHz? > > Thanks & Regards, > Brendan. > > _______________________________________________ > USRP-users mailing list -- [email protected] > To unsubscribe send an email to [email protected]
_______________________________________________ USRP-users mailing list -- [email protected] To unsubscribe send an email to [email protected]
