Dear Ivan,
On Sat, 2019-05-11 at 20:00 +0300, Ivan Zahartchuk wrote:
> Sorry I did not specify. When working with the start_cont mode with a
> frequency of more than 5 MHz, I have an overflow error.
Which probably happens due to the inefficient way you handle the data;
your program simply is too slow.
> That leads to a chaotic change in the spectrum. I agree about the
> wrong allocation of memory, but I created my own data array and ran
> it through the whole chain of changes and additions and had what I
> expected at the output.
Huh? When using offline data, your computationally load doesn't matter;
I'm not sure I'm understanding you correctly here!
> Moreover, the data array was much more than what I get from the
> board. In this regard, I decided that the matter is in the reading
> from the board.
I'll allow myself to doubt that: Again, the fact that your data array
runs through smoothly is no indication for your software being fast
enough to keep up with the torrent of samples UHD will give you! When
it doesn't keep up, you see exactly what you're describing: UHD
reporting overflows due to your software not fetching samples fast
enough. Overflows aren't UHD bugs, they are problems in the way you use
UHD!
> I have seen examples, but start_cont mode is used there and there
> are not so many frequency adjustments. Therefore, I decided to use
> the num_sams_and_done mode, but there is also a problem with it which
> is described in the previous message. If you have more suggestions, I
> will be very grateful to you.
Well, exactly what I wrote in my previous mail: you need to make your
software fast enough. Preallocate the buffer; don't append to its end.
Aside from that, do some profiling. Under linux, `perf top -ag python
yourscript.py` is your friend to figure out where your program spends
its CPU cycles.
Best regards,
Marcus
>
> сб, 11 мая 2019 г. в 22:27, Marcus Müller <marcus.muel...@ettus.com>:
> > Dear Ivan,
> >
> > it's not clear what you've modified. I'm not aware of any UHD
> > function
> > that restricts any frequency to 5 MHz.
> > Could you elaborate on which code you're basing this on?
> >
> > Also, while I really like the Python interface, if you're really
> > after
> > high-rate sampling, it might simply not be the optimal thing to
> > use:
> > You'd have to be very careful in Pythonland to not run into
> > performance
> > problems once you've gotten the samples from UHD:
> >
> > complex_buffs=np.append(complex_buffs,result).ravel()
> >
> > A very bad idea. You're constantly re-allocating buffers here.
> > Don't do
> > that. No matter in which programming language you'd do this, you'd
> > make
> > sure that the buffer is large enough for your data collection to
> > begin
> > with and then tell the UHD library to fill the appropriate part in
> > that
> > buffer to avoid a) having to ask for a larger buffer regularly and
> > b)
> > avoid copying data.
> > Asking for an appended version of your last buffer means that
> > something
> > has to allocate a larger buffer – which comes at very large cost!
> >
> > Best regards,
> > Marcus
> >
> > On Sat, 2019-05-11 at 21:31 +0300, Ivan Zahartchuk via USRP-users
> > wrote:
> > > Hello. My task is to make a broadband spectrum analyzer on the
> > usrp
> > > b200 board. For this, the standard functions for reading samples
> > in
> > > python are not suitable for me. Therefore, I edited them. When
> > > reading samples using the start_con method, I cannot specify a
> > band
> > > greater than 5 MHz. Therefore, I use the num_sams_and_done
> > method.
> > > But I have problems with him. The fact is that my frequency which
> > I
> > > know appears in the wrong place. For example, I generate a
> > frequency
> > > of 910 MHz and it appears at 930 MHz (with a bandwidth of 20
> > MHz). I
> > > can not understand what caused it. Here are my reading functions
> > in
> > > two ways. I would be extremely grateful for the help.
> > >
> > >
> > >
> > >
> > > def test_reciev(self,freq,bandwich):
> > > complex_buffs=np.array([])
> > > buffs = np.array([])
> > > result = np.empty((len([0]), self.samples),
> > dtype=np.complex64)
> > >
> > > for i, freqq in enumerate(freq):
> > >
> > > recv_samps = 0
> > > #self.usrp.set_rx_rate(bandwich[i])
> > > k=uhd.types.TuneRequest(freqq)
> > > #k.args(uhd.types.device_addr("mode_n=integer"))
> > > self.usrp.set_rx_freq(k)
> > > self.stream_cmd =
> > > uhd.types.StreamCMD(uhd.types.StreamMode.start_cont)
> > > self.stream_cmd.stream_now = True
> > > self.streamer_rx.issue_stream_cmd(self.stream_cmd)
> > > while self.usrp.get_rx_sensor("lo_locked").to_bool() !=
> > True:
> > > continue
> > >
> > > samps = np.array([], dtype=np.complex64)
> > > while recv_samps < self.samples:
> > >
> > > samps = self.streamer_rx.recv(self.recv_buff,
> > > self.metadata_rx)
> > > if self.metadata_rx.error_code !=
> > > lib.types.rx_metadata_error_code.none:
> > > print(self.metadata_rx.strerror())
> > > if samps:
> > > real_samps = min(self.samples - recv_samps,
> > samps)
> > > result[:, recv_samps:recv_samps + real_samps] =
> > > self.recv_buff[:, 0:real_samps]
> > > recv_samps += real_samps
> > > #print (self.usrp.get_rx_sensor('rssi'))
> > > #print(self.streamer_rx.get_max_num_samps())
> > > #while samps:
> > > # samps = self.streamer_rx.recv(self.recv_buff,
> > > self.metadata_rx)
> > >
> > >
> > > #self.stream_cmd.time_spec = lib.types.time_spec(0)
> > > self.stream_cmd =
> > > lib.types.stream_cmd(lib.types.stream_mode.stop_cont)
> > > self.streamer_rx.issue_stream_cmd(self.stream_cmd)
> > >
> > > complex_buffs=np.append(complex_buffs,result).ravel()
> > > #correct_result=result
> > > correct_result_1=result-
> > > (np.mean(result.real)+np.mean(result.imag)*1j)
> > > #correct_result.real=result.real-np.mean(result.real)
> > > #correct_result.imag = result.imag - np.mean(result.imag)
> > >
> > > PSD = self.fft_test(result)
> > >
> > >
> > > #PSD[8188:8202]=np.mean(PSD[8180:8188])
> > >
> > >
> > > buffs = np.append(buffs,PSD)
> > >
> > >
> > >
> > > return complex_buffs,
> > > buffs#np.append(buffs[value.samples:],buffs[:value.samples])
> > >
> > >
> > >
> > > def test_reciev(self,freq,bandwich):
> > > complex_buffs=np.array([])
> > > buffs = np.array([])
> > > result = np.empty((len([0]), self.samples),
> > dtype=np.complex64)
> > >
> > > for i, freqq in enumerate(freq):
> > >
> > > recv_samps = 0
> > > #self.usrp.set_rx_rate(bandwich[i])
> > > k=uhd.types.TuneRequest(freqq)
> > > #k.args(uhd.types.device_addr("mode_n=integer"))
> > > self.usrp.set_rx_freq(k)
> > >
> > > while self.usrp.get_rx_sensor("lo_locked").to_bool() !=
> > True:
> > > continue
> > >
> > >
> > > while recv_samps < self.samples:
> > >
> > > samps = self.streamer_rx.recv(self.recv_buff,
> > > self.metadata_rx)
> > > if self.metadata_rx.error_code !=
> > > lib.types.rx_metadata_error_code.none:
> > > print(self.metadata_rx.strerror())
> > > if samps:
> > > real_samps = min(self.samples - recv_samps,
> > samps)
> > > result[:, recv_samps:recv_samps + real_samps] =
> > > self.recv_buff[:, 0:real_samps]
> > > recv_samps += real_samps
> > > #print (self.usrp.get_rx_sensor('rssi'))
> > >
> > >
> > > self.stream_cmd.time_spec = lib.types.time_spec(0)
> > > self.streamer_rx.issue_stream_cmd(self.stream_cmd)
> > >
> > > complex_buffs=np.append(complex_buffs,result).ravel()
> > > correct_result=result
> > > correct_result_1=result-
> > > (np.mean(result.real)+np.mean(result.imag)*1j)
> > > correct_result.real=result.real-np.mean(result.real)
> > > correct_result.imag = result.imag - np.mean(result.imag)
> > >
> > > PSD = self.fft_test(result)
> > >
> > >
> > > #PSD[8188:8202]=np.mean(PSD[8180:8188])
> > >
> > >
> > > buffs = np.append(buffs,PSD)
> > >
> > >
> > >
> > > return complex_buffs,
> > > buffs#np.append(buffs[value.samples:],buffs[:value.samples])
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > > _______________________________________________
> > > 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
