Hi folks, I was thinking if it is possible to sample with more than 50 MSps (i.e. 100 MSps) on the N210. I know the Gbit Ethernet is the bottleneck, but if I want to transmit only short bursts, and receive short bursts, it won't be a bottleneck anymore. For example, in a TDD mode, where I transmit 50 % of the time and I receive 50 % of the time, with 8 bit samples, and sample rate of 100 MSps, the Ethernet throughput would be exactly 1 Gbps in the both directions. Is this supported by UHD and the firmware? BR, Vladica

Hi Vladica, out of the box, this will be hard; I haven't checked if this actually works, but I slightly doubt it – the limiting factor is that you can't pre-buffer that much on an N210, and that would limit the durations of your bursts severely. On a different note: what's your use case? Aside from the filterless BasicRX/TX, all daughterboards for the N210 have at most 40 MHz of analog bandwidth - hence, your undecimated 100MS/s stream would carry no more information than the same stream at 50MS/s. Best regards, Marcus On 23.06.2016 11:04, Vladica Sark via USRP-users wrote: > Hi folks, > > I was thinking if it is possible to sample with more than 50 MSps > (i.e. 100 MSps) on the N210. I know the Gbit Ethernet is the > bottleneck, but if I want to transmit only short bursts, and receive > short bursts, it won't be a bottleneck anymore. For example, in a TDD > mode, where I transmit 50 % of the time and I receive 50 % of the > time, with 8 bit samples, and sample rate of 100 MSps, the Ethernet > throughput would be exactly 1 Gbps in the both directions. > > Is this supported by UHD and the firmware? > > BR, > Vladica > > _______________________________________________ > USRP-users mailing list > USRP-users@lists.ettus.com > http://lists.ettus.com/mailman/listinfo/usrp-users_lists.ettus.com

Hi Marcus, In my case data would be streamed not more than 20% of the time. It is a TDM scenario, where the radios have only a small slots for transmission i.e. reception. Therefore, basically, the rate would be way below 1 Gbps. Regarding the bandwidth, I would like to do some oversampling of the signal (2x) or so, in order do make better timing recovery. It is also an application where RF ranging between the stations is to be performed. In this cases, a nanosecond timing should be measured. Basically, perfect reconstruction can be made with sinc functions, but that is practically not possible. I agree that even interpolation, for reconstruction of a higher sample rate signal, would introduce only small errors, but they matter in RF ranging applications. Therefore, sampling with higher sample rate, would be required to minimize latter the timing errors, when the time of arrival is to be estimated. BR, Vladica On 23.06.2016 11:21, Marcus Müller via USRP-users wrote: > Hi Vladica, > > out of the box, this will be hard; I haven't checked if this actually > works, but I slightly doubt it – the limiting factor is that you can't > pre-buffer that much on an N210, and that would limit the durations of > your bursts severely. > > On a different note: what's your use case? Aside from the filterless > BasicRX/TX, all daughterboards for the N210 have at most 40 MHz of > analog bandwidth - hence, your undecimated 100MS/s stream would carry no > more information than the same stream at 50MS/s. > > Best regards, > Marcus > > On 23.06.2016 11:04, Vladica Sark via USRP-users wrote: >> Hi folks, >> >> I was thinking if it is possible to sample with more than 50 MSps >> (i.e. 100 MSps) on the N210. I know the Gbit Ethernet is the >> bottleneck, but if I want to transmit only short bursts, and receive >> short bursts, it won't be a bottleneck anymore. For example, in a TDD >> mode, where I transmit 50 % of the time and I receive 50 % of the >> time, with 8 bit samples, and sample rate of 100 MSps, the Ethernet >> throughput would be exactly 1 Gbps in the both directions. >> >> Is this supported by UHD and the firmware? >> >> BR, >> Vladica >> >> _______________________________________________ >> USRP-users mailing list >> USRP-users@lists.ettus.com >> http://lists.ettus.com/mailman/listinfo/usrp-users_lists.ettus.com > > > _______________________________________________ > USRP-users mailing list > USRP-users@lists.ettus.com > http://lists.ettus.com/mailman/listinfo/usrp-users_lists.ettus.com >

An BTW, is there a way to see the state of the buffer memory in the N210, before putting a new transmit burst in it? Just not to overfill it. BR, Vladica On 23.06.2016 11:21, Marcus Müller via USRP-users wrote: > Hi Vladica, > > out of the box, this will be hard; I haven't checked if this actually > works, but I slightly doubt it – the limiting factor is that you can't > pre-buffer that much on an N210, and that would limit the durations of > your bursts severely. > > On a different note: what's your use case? Aside from the filterless > BasicRX/TX, all daughterboards for the N210 have at most 40 MHz of > analog bandwidth - hence, your undecimated 100MS/s stream would carry no > more information than the same stream at 50MS/s. > > Best regards, > Marcus > > On 23.06.2016 11:04, Vladica Sark via USRP-users wrote: >> Hi folks, >> >> I was thinking if it is possible to sample with more than 50 MSps >> (i.e. 100 MSps) on the N210. I know the Gbit Ethernet is the >> bottleneck, but if I want to transmit only short bursts, and receive >> short bursts, it won't be a bottleneck anymore. For example, in a TDD >> mode, where I transmit 50 % of the time and I receive 50 % of the >> time, with 8 bit samples, and sample rate of 100 MSps, the Ethernet >> throughput would be exactly 1 Gbps in the both directions. >> >> Is this supported by UHD and the firmware? >> >> BR, >> Vladica >> >> _______________________________________________ >> USRP-users mailing list >> USRP-users@lists.ettus.com >> http://lists.ettus.com/mailman/listinfo/usrp-users_lists.ettus.com > > > _______________________________________________ > USRP-users mailing list > USRP-users@lists.ettus.com > http://lists.ettus.com/mailman/listinfo/usrp-users_lists.ettus.com >

Hi Vladica, my radar knowledge might be a bit rusty here, but wouldn't ranging resolution be only proportional to effective signal bandwidth – and that being limited by the analog bandwidth, here? Regarding the technical feasibility: With a minor hack in tx_dsp_core_200.cpp::get_host_rates to allow rates down to decimation==1 (even if these exceed the link rate), you should be able to bypass all decimating filters on the USRP, and just try. In essence, you'd send out finite acquisition rx stream commands with appropriate timespecs for your tdd RX, and and fill in the correct tx_metadata for send() for TX. You can't overfill the on-device buffers. The USRP will simply not ack further packets if the buffers are full, leading to back-pressure on the host. Best regards, Marcus On 23.06.2016 11:42, Vladica Sark via USRP-users wrote: > Hi Marcus, > > In my case data would be streamed not more than 20% of the time. It is > a TDM scenario, where the radios have only a small slots for > transmission i.e. reception. Therefore, basically, the rate would be > way below 1 Gbps. > > Regarding the bandwidth, I would like to do some oversampling of the > signal (2x) or so, in order do make better timing recovery. It is also > an application where RF ranging between the stations is to be > performed. In this cases, a nanosecond timing should be measured. > Basically, perfect reconstruction can be made with sinc functions, but > that is practically not possible. I agree that even interpolation, for > reconstruction of a higher sample rate signal, would introduce only > small errors, but they matter in RF ranging applications. Therefore, > sampling with higher sample rate, would be required to minimize latter > the timing errors, when the time of arrival is to be estimated. > > > BR, > Vladica > > > > On 23.06.2016 11:21, Marcus Müller via USRP-users wrote: >> Hi Vladica, >> >> out of the box, this will be hard; I haven't checked if this actually >> works, but I slightly doubt it – the limiting factor is that you can't >> pre-buffer that much on an N210, and that would limit the durations of >> your bursts severely. >> >> On a different note: what's your use case? Aside from the filterless >> BasicRX/TX, all daughterboards for the N210 have at most 40 MHz of >> analog bandwidth - hence, your undecimated 100MS/s stream would carry no >> more information than the same stream at 50MS/s. >> >> Best regards, >> Marcus >> >> On 23.06.2016 11:04, Vladica Sark via USRP-users wrote: >>> Hi folks, >>> >>> I was thinking if it is possible to sample with more than 50 MSps >>> (i.e. 100 MSps) on the N210. I know the Gbit Ethernet is the >>> bottleneck, but if I want to transmit only short bursts, and receive >>> short bursts, it won't be a bottleneck anymore. For example, in a TDD >>> mode, where I transmit 50 % of the time and I receive 50 % of the >>> time, with 8 bit samples, and sample rate of 100 MSps, the Ethernet >>> throughput would be exactly 1 Gbps in the both directions. >>> >>> Is this supported by UHD and the firmware? >>> >>> BR, >>> Vladica >>> >>> _______________________________________________ >>> USRP-users mailing list >>> USRP-users@lists.ettus.com >>> http://lists.ettus.com/mailman/listinfo/usrp-users_lists.ettus.com >> >> >> _______________________________________________ >> USRP-users mailing list >> USRP-users@lists.ettus.com >> http://lists.ettus.com/mailman/listinfo/usrp-users_lists.ettus.com >> > > _______________________________________________ > USRP-users mailing list > USRP-users@lists.ettus.com > http://lists.ettus.com/mailman/listinfo/usrp-users_lists.ettus.com

Hi Markus, You are completely right about the radars. Yes, the BW and SNR are the main players. Anyway, my issue is with the estimator. In my case, the estimator is sensitive, so to say, on the sample rate. There are other approaches, where this can be avoided, but then another issues arise. Thanks for the hack. I would try it soon, to see what comes out. The approach is exactly the one you described, acquisition of finite number of samples with a given timespecs and transmition of finite length frames with given timespecs. BR, Vladica On 23.06.2016 11:54, Marcus Müller via USRP-users wrote: > Hi Vladica, > > my radar knowledge might be a bit rusty here, but wouldn't ranging > resolution be only proportional to effective signal bandwidth – and that > being limited by the analog bandwidth, here? > > Regarding the technical feasibility: With a minor hack in > tx_dsp_core_200.cpp::get_host_rates to allow rates down to decimation==1 > (even if these exceed the link rate), you should be able to bypass all > decimating filters on the USRP, and just try. In essence, you'd send out > finite acquisition rx stream commands with appropriate timespecs for > your tdd RX, and and fill in the correct tx_metadata for send() for TX. > > You can't overfill the on-device buffers. The USRP will simply not ack > further packets if the buffers are full, leading to back-pressure on the > host. > > Best regards, > Marcus > > On 23.06.2016 11:42, Vladica Sark via USRP-users wrote: >> Hi Marcus, >> >> In my case data would be streamed not more than 20% of the time. It is >> a TDM scenario, where the radios have only a small slots for >> transmission i.e. reception. Therefore, basically, the rate would be >> way below 1 Gbps. >> >> Regarding the bandwidth, I would like to do some oversampling of the >> signal (2x) or so, in order do make better timing recovery. It is also >> an application where RF ranging between the stations is to be >> performed. In this cases, a nanosecond timing should be measured. >> Basically, perfect reconstruction can be made with sinc functions, but >> that is practically not possible. I agree that even interpolation, for >> reconstruction of a higher sample rate signal, would introduce only >> small errors, but they matter in RF ranging applications. Therefore, >> sampling with higher sample rate, would be required to minimize latter >> the timing errors, when the time of arrival is to be estimated. >> >> >> BR, >> Vladica >> >> >> >> On 23.06.2016 11:21, Marcus Müller via USRP-users wrote: >>> Hi Vladica, >>> >>> out of the box, this will be hard; I haven't checked if this actually >>> works, but I slightly doubt it – the limiting factor is that you can't >>> pre-buffer that much on an N210, and that would limit the durations of >>> your bursts severely. >>> >>> On a different note: what's your use case? Aside from the filterless >>> BasicRX/TX, all daughterboards for the N210 have at most 40 MHz of >>> analog bandwidth - hence, your undecimated 100MS/s stream would carry no >>> more information than the same stream at 50MS/s. >>> >>> Best regards, >>> Marcus >>> >>> On 23.06.2016 11:04, Vladica Sark via USRP-users wrote: >>>> Hi folks, >>>> >>>> I was thinking if it is possible to sample with more than 50 MSps >>>> (i.e. 100 MSps) on the N210. I know the Gbit Ethernet is the >>>> bottleneck, but if I want to transmit only short bursts, and receive >>>> short bursts, it won't be a bottleneck anymore. For example, in a TDD >>>> mode, where I transmit 50 % of the time and I receive 50 % of the >>>> time, with 8 bit samples, and sample rate of 100 MSps, the Ethernet >>>> throughput would be exactly 1 Gbps in the both directions. >>>> >>>> Is this supported by UHD and the firmware? >>>> >>>> BR, >>>> Vladica >>>> >>>> _______________________________________________ >>>> USRP-users mailing list >>>> USRP-users@lists.ettus.com >>>> http://lists.ettus.com/mailman/listinfo/usrp-users_lists.ettus.com >>> >>> >>> _______________________________________________ >>> USRP-users mailing list >>> USRP-users@lists.ettus.com >>> http://lists.ettus.com/mailman/listinfo/usrp-users_lists.ettus.com >>> >> >> _______________________________________________ >> USRP-users mailing list >> USRP-users@lists.ettus.com >> http://lists.ettus.com/mailman/listinfo/usrp-users_lists.ettus.com > > > _______________________________________________ > USRP-users mailing list > USRP-users@lists.ettus.com > http://lists.ettus.com/mailman/listinfo/usrp-users_lists.ettus.com >

See the recent thread titled 'Size of TX Buffer on N210'. M On 06/23/2016 02:50 AM, Vladica Sark via USRP-users wrote: > An BTW, is there a way to see the state of the buffer memory in the > N210, before putting a new transmit burst in it? Just not to overfill it. > > BR, > Vladica > > > On 23.06.2016 11:21, Marcus Müller via USRP-users wrote: >> Hi Vladica, >> >> out of the box, this will be hard; I haven't checked if this actually >> works, but I slightly doubt it – the limiting factor is that you can't >> pre-buffer that much on an N210, and that would limit the durations of >> your bursts severely. >> >> On a different note: what's your use case? Aside from the filterless >> BasicRX/TX, all daughterboards for the N210 have at most 40 MHz of >> analog bandwidth - hence, your undecimated 100MS/s stream would carry no >> more information than the same stream at 50MS/s. >> >> Best regards, >> Marcus >> >> On 23.06.2016 11:04, Vladica Sark via USRP-users wrote: >>> Hi folks, >>> >>> I was thinking if it is possible to sample with more than 50 MSps >>> (i.e. 100 MSps) on the N210. I know the Gbit Ethernet is the >>> bottleneck, but if I want to transmit only short bursts, and receive >>> short bursts, it won't be a bottleneck anymore. For example, in a TDD >>> mode, where I transmit 50 % of the time and I receive 50 % of the >>> time, with 8 bit samples, and sample rate of 100 MSps, the Ethernet >>> throughput would be exactly 1 Gbps in the both directions. >>> >>> Is this supported by UHD and the firmware? >>> >>> BR, >>> Vladica >>> >>> _______________________________________________ >>> USRP-users mailing list >>> USRP-users@lists.ettus.com >>> http://lists.ettus.com/mailman/listinfo/usrp-users_lists.ettus.com >> >> >> _______________________________________________ >> USRP-users mailing list >> USRP-users@lists.ettus.com >> http://lists.ettus.com/mailman/listinfo/usrp-users_lists.ettus.com >> > > _______________________________________________ > USRP-users mailing list > USRP-users@lists.ettus.com > http://lists.ettus.com/mailman/listinfo/usrp-users_lists.ettus.com