usrp-users@lists.ettus.com
Discussion and technical support related to USRP, UHD, RFNoC
View all threadsRe: [USRP-users] N210+SBX40MHz: Transfer Function of RF path?
Stephan,
With reference to the following diagram:
http://ianbuckley.net/USRP_Advanced_Diagram_annotated.pdf
and to this nice picture of the various filter characteristics plotted
against normalized frequency:
http://witestlab.poly.edu/~ffund/el9043/labs/images/lab2-filters.png
For the N200/N210/USRP2, at all versions of UHD and also the B200/B210 with
UHD of 3.8.5 or earlier.
The N2x0 has a fixed sample rate of 100MHz at the interface to the discrete
DAC/ADC's used (The B2x0's is programmable <61.44MHz)...at Ettus research
this value is termed the "master_clock_rate".
The CORDIC and CIC filter operate at this sample rate always. The CIC
provides flexible programmable decimation/interpolation for all integer
ratios between 1:1 and 1:128 but a non linear passband and poor roll-off
into the stopband.
The 7 tap FIR implements a halfband lowpass filter with fixed
decimation/interpolation of 2 and can also run at upto the
master_clock_rate rate.
The 31 tap FIR implements a halfband lowpass filter with fixed
decimation/interpolation of 2 (superior to the 7 tap one) but can only run
upto half of the master_clock_rate.
UHD will always try to use the best combination of filters to satisfy a
requested decimation/interpolation rate.
Thus for decimation/interpolating values that are odd numbers, only the CIC
filter will (can) be used.
For a decimation/interpolation value of exactly 2 only the 7 tap FIR will
(can) be used.
For decimation/interpolation values > 2 that factor by 2 (but not 4) then
the CIC will be used in conjunction with the 31tap FIR.
For decimation/interpolation values >=4 that also factor by 4 then the CIC
will be used in conjunction with both 7 and 31 tap FIR's.
Thus from a system design perspective its always desirable to use a sample
rate N, where:
(master_clock_rate / 4) / N = integer value (for example for N210:
25MHz,12.5MHz, 6.25MHz, 8.333MHz...)
since this gives a largely linear response through most of the passband. If
the application demands a sample rate thats not related to the
master_clock_rate in a simple integer ratio, then the best strategy is to
reduce the sample rate as much as possible within the USRP then complete
the sample rate conversion on the Host with another filter implemented in
S/W.
Side notes:
- Could one of the AE team write this up better and include it in the on
line notes pls as it gets asked repeatedly. - Diagrams (miss) appropriated from:
http://witestlab.poly.edu/~ffund/el9043/labs/lab2.html
http://www.ni.com/white-paper/14311/en/
-Ian
On Mon, Feb 22, 2016 at 11:28 PM, Ludwig Stephan (CR/AEH4) <
Stephan.Ludwig2@de.bosch.com> wrote:
Hi Ian,
thank you very much, your explanation helped me a lot. Using 12.5 MHz of
sample rate makes the channel flat, in deed.
Could you provide any pointer to some information about the criteria that
lead to the different filter parameters, please? Just in order to avoid any
further complications if I need to scale sample frequency again.
Thanks again for your valuable help.
Best regards
Stephan Ludwig
Communication Technology (CR/AEH4)
Robert Bosch GmbH | Renningen | 70465 Stuttgart | GERMANY | www.bosch.com
Tel. +49(711)811-8809 | Mobile +49(172)5630639 | Fax +49(711)811-5187845
| Stephan.Ludwig2@de.bosch.com
Registered Office: Stuttgart, Registration Court: Amtsgericht Stuttgart,
HRB 14000;
Chairman of the Supervisory Board: Franz Fehrenbach; Managing Directors:
Dr. Volkmar Denner,
Dr. Stefan Asenkerschbaumer, Dr. Rolf Bulander, Dr. Stefan Hartung, Dr.
Markus Heyn, Dr. Dirk Hoheisel,
Christoph Kübel, Uwe Raschke, Dr. Werner Struth, Peter Tyroller
-----Ursprüngliche Nachricht-----
Von: Ian Buckley [mailto:ianb@ionconcepts.com]
Gesendet: Freitag, 19. Februar 2016 02:41
An: Ludwig Stephan (CR/AEH4) Stephan.Ludwig2@de.bosch.com
Cc: usrp-users@lists.ettus.com USRP-users usrp-users@lists.ettus.com;
support@ettus.com
Betreff: Re: [USRP-users] N210+SBX40MHz: Transfer Function of RF path?
Stephan,
If I understand you correctly, you are decimating by 10 in the N210 to get
10 MSamples/Sec before passing data to GNURadio, correct? Possibly you are
also doing this on the transmit side also?
This will result in a less than ideal digital filter
configuration.basically a CIC filter cascaded into a 7tap half band FIR
filter. The CIC is approximately flat for Fs/4 in the band center before it
starts to roll off significantly. The FIR is also not very flat beyond
about ~0.35Fs where the stop band roll off starts.
I suggest you repeat you experiment using 12.5MSamples/Sec instead. This
will result in a cascade of CIC into 7tap FIR into 31tap FIR. This
configuration should be very flat out to about ~Fs0.45.
You will never see a flat response across the whole bandwidth because real
world practical filters have to have start rolling off before the band edge
to give a reasonable stop band rejection to prevent aliasing from adjacent
frequencies.
-Ian
On Feb 18, 2016, at 8:29 AM, Ludwig Stephan (CR/AEH4) via USRP-users <
usrp-users@lists.ettus.com> wrote:
Hello List,
I am using a 2 x USRP N210 + SBX40MHz daughter board for measuring
channels. I use a OFDM (802.11g) transmission signal @2.45 GHz carrier
frequency for estimating the channel in frequency domain. Everything works
fine, I even receive data packets without error. But when I use a coax
cable between Tx and Rx, I do not obtain a flat transfer function as I
expect (cp. Attached image). From the center frequency, the magnitude drops
by some 8 dB at the band edge. This is also true for a short (2m distance)
over-the-air transmission in an office scenario, where we also expect a
flat channel, because the echos all fall within the first tap. Du to my
limited processing power, I scaled the transmit signal to 10 MHz bandwidth
(each OFDM symbol then has double the length) instead of 20 MHz by changing
the sample rate accordingly, leaving the GNU Radio DSP unchanged.
I checked with different hardware (we have 4 USRPs) in different Tx/Rx
combinations. When I further decrease the used bandwidth/sample rate, I get
the a similar transfer function but only the corresponding portion of the
one at 10 MHz band width - in the picture the share from -2.5 MHz to +2.5
MHz at same magnitude. Thus the problem is not in the signal processing
done in GNU radio, but must be either in the DSP on FPGA/N210 or in the RF
path of the SBX. I cannot imagine that such a large frequency selectivity
in DSP/RF hardware is wanted. The SBX is specified for 40 MHz bandwith. And
I doubt that this is a sinc D/A-C distortion, because 10 MHz is a tenth of
the D/A-C sample rate.
Could you give me some hint for an explanation, please? Any help in
changing this?
Best regards
Stephan Ludwig
Communication Technology (CR/AEH4)
Robert Bosch GmbH | Renningen | 70465 Stuttgart | GERMANY |
www.bosch.com Tel. +49(711)811-8809 | Mobile +49(172)5630639 | Fax
+49(711)811-5187845 | Stephan.Ludwig2@de.bosch.com
Registered Office: Stuttgart, Registration Court: Amtsgericht
Stuttgart, HRB 14000; Chairman of the Supervisory Board: Franz
Fehrenbach; Managing Directors: Dr. Volkmar Denner, Dr. Stefan
Asenkerschbaumer, Dr. Rolf Bulander, Dr. Stefan Hartung, Dr. Markus
Heyn, Dr. Dirk Hoheisel, Christoph Kübel, Uwe Raschke, Dr. Werner
Struth, Peter Tyroller
<CableTransfer.png>_______________________________________________
USRP-users mailing list
USRP-users@lists.ettus.com
http://lists.ettus.com/mailman/listinfo/usrp-users_lists.ettus.com
Hi Ian,
thanks a lot for this comprehensive and excellent explanation. That really helps me a lot. This completely solves my problem.
+1 for adding to online doc
Best regards
Stephan Ludwig
Communication Technology (CR/AEH4)
Robert Bosch GmbH | Renningen | 70465 Stuttgart | GERMANY | www.bosch.comhttp://www.bosch.com
Tel. +49(711)811-8809 | Mobile +49(172)5630639 | Fax +49(711)811-5187845 | Stephan.Ludwig2@de.bosch.commailto:Stephan.Ludwig2@de.bosch.com
Registered Office: Stuttgart, Registration Court: Amtsgericht Stuttgart, HRB 14000;
Chairman of the Supervisory Board: Franz Fehrenbach; Managing Directors: Dr. Volkmar Denner,
Dr. Stefan Asenkerschbaumer, Dr. Rolf Bulander, Dr. Stefan Hartung, Dr. Markus Heyn, Dr. Dirk Hoheisel,
Christoph Kübel, Uwe Raschke, Dr. Werner Struth, Peter Tyroller
Von: Ian Buckley [mailto:ianb@ionconcepts.com]
Gesendet: Dienstag, 23. Februar 2016 15:21
An: Ludwig Stephan (CR/AEH4) Stephan.Ludwig2@de.bosch.com; USRP-users@lists.ettus.com; Ettus Research Support support@ettus.com
Betreff: Re: [USRP-users] N210+SBX40MHz: Transfer Function of RF path?
Stephan,
With reference to the following diagram:
http://ianbuckley.net/USRP_Advanced_Diagram_annotated.pdf
and to this nice picture of the various filter characteristics plotted against normalized frequency:
http://witestlab.poly.edu/~ffund/el9043/labs/images/lab2-filters.png
For the N200/N210/USRP2, at all versions of UHD and also the B200/B210 with UHD of 3.8.5 or earlier.
The N2x0 has a fixed sample rate of 100MHz at the interface to the discrete DAC/ADC's used (The B2x0's is programmable <61.44MHz)...at Ettus research this value is termed the "master_clock_rate".
The CORDIC and CIC filter operate at this sample rate always. The CIC provides flexible programmable decimation/interpolation for all integer ratios between 1:1 and 1:128 but a non linear passband and poor roll-off into the stopband.
The 7 tap FIR implements a halfband lowpass filter with fixed decimation/interpolation of 2 and can also run at upto the master_clock_rate rate.
The 31 tap FIR implements a halfband lowpass filter with fixed decimation/interpolation of 2 (superior to the 7 tap one) but can only run upto half of the master_clock_rate.
UHD will always try to use the best combination of filters to satisfy a requested decimation/interpolation rate.
Thus for decimation/interpolating values that are odd numbers, only the CIC filter will (can) be used.
For a decimation/interpolation value of exactly 2 only the 7 tap FIR will (can) be used.
For decimation/interpolation values > 2 that factor by 2 (but not 4) then the CIC will be used in conjunction with the 31tap FIR.
For decimation/interpolation values >=4 that also factor by 4 then the CIC will be used in conjunction with both 7 and 31 tap FIR's.
Thus from a system design perspective its always desirable to use a sample rate N, where:
(master_clock_rate / 4) / N = integer value (for example for N210: 25MHz,12.5MHz, 6.25MHz, 8.333MHz...)
since this gives a largely linear response through most of the passband. If the application demands a sample rate thats not related to the master_clock_rate in a simple integer ratio, then the best strategy is to reduce the sample rate as much as possible within the USRP then complete the sample rate conversion on the Host with another filter implemented in S/W.
Side notes:
- Could one of the AE team write this up better and include it in the on line notes pls as it gets asked repeatedly.
- Diagrams (miss) appropriated from:
http://witestlab.poly.edu/~ffund/el9043/labs/lab2.html
http://www.ni.com/white-paper/14311/en/
-Ian
On Mon, Feb 22, 2016 at 11:28 PM, Ludwig Stephan (CR/AEH4) <Stephan.Ludwig2@de.bosch.commailto:Stephan.Ludwig2@de.bosch.com> wrote:
Hi Ian,
thank you very much, your explanation helped me a lot. Using 12.5 MHz of sample rate makes the channel flat, in deed.
Could you provide any pointer to some information about the criteria that lead to the different filter parameters, please? Just in order to avoid any further complications if I need to scale sample frequency again.
Thanks again for your valuable help.
Best regards
Stephan Ludwig
Communication Technology (CR/AEH4)
Robert Bosch GmbH | Renningen | 70465 Stuttgart | GERMANY | www.bosch.comhttp://www.bosch.com
Tel. +49(711)811-8809tel:%2B49%28711%29811-8809 | Mobile +49(172)5630639tel:%2B49%28172%295630639 | Fax +49(711)811-5187845tel:%2B49%28711%29811-5187845 | Stephan.Ludwig2@de.bosch.commailto:Stephan.Ludwig2@de.bosch.com
Registered Office: Stuttgart, Registration Court: Amtsgericht Stuttgart, HRB 14000;
Chairman of the Supervisory Board: Franz Fehrenbach; Managing Directors: Dr. Volkmar Denner,
Dr. Stefan Asenkerschbaumer, Dr. Rolf Bulander, Dr. Stefan Hartung, Dr. Markus Heyn, Dr. Dirk Hoheisel,
Christoph Kübel, Uwe Raschke, Dr. Werner Struth, Peter Tyroller
-----Ursprüngliche Nachricht-----
Von: Ian Buckley [mailto:ianb@ionconcepts.commailto:ianb@ionconcepts.com]
Gesendet: Freitag, 19. Februar 2016 02:41
An: Ludwig Stephan (CR/AEH4) <Stephan.Ludwig2@de.bosch.commailto:Stephan.Ludwig2@de.bosch.com>
Cc: usrp-users@lists.ettus.commailto:usrp-users@lists.ettus.com USRP-users <usrp-users@lists.ettus.commailto:usrp-users@lists.ettus.com>; support@ettus.commailto:support@ettus.com
Betreff: Re: [USRP-users] N210+SBX40MHz: Transfer Function of RF path?
Stephan,
If I understand you correctly, you are decimating by 10 in the N210 to get 10 MSamples/Sec before passing data to GNURadio, correct? Possibly you are also doing this on the transmit side also?
This will result in a less than ideal digital filter configuration.basically a CIC filter cascaded into a 7tap half band FIR filter. The CIC is approximately flat for Fs/4 in the band center before it starts to roll off significantly. The FIR is also not very flat beyond about ~0.35Fs where the stop band roll off starts.
I suggest you repeat you experiment using 12.5MSamples/Sec instead. This will result in a cascade of CIC into 7tap FIR into 31tap FIR. This configuration should be very flat out to about ~Fs0.45.
You will never see a flat response across the whole bandwidth because real world practical filters have to have start rolling off before the band edge to give a reasonable stop band rejection to prevent aliasing from adjacent frequencies.
-Ian
On Feb 18, 2016, at 8:29 AM, Ludwig Stephan (CR/AEH4) via USRP-users <usrp-users@lists.ettus.commailto:usrp-users@lists.ettus.com> wrote:
Hello List,
I am using a 2 x USRP N210 + SBX40MHz daughter board for measuring channels. I use a OFDM (802.11g) transmission signal @2.45 GHz carrier frequency for estimating the channel in frequency domain. Everything works fine, I even receive data packets without error. But when I use a coax cable between Tx and Rx, I do not obtain a flat transfer function as I expect (cp. Attached image). From the center frequency, the magnitude drops by some 8 dB at the band edge. This is also true for a short (2m distance) over-the-air transmission in an office scenario, where we also expect a flat channel, because the echos all fall within the first tap. Du to my limited processing power, I scaled the transmit signal to 10 MHz bandwidth (each OFDM symbol then has double the length) instead of 20 MHz by changing the sample rate accordingly, leaving the GNU Radio DSP unchanged.
I checked with different hardware (we have 4 USRPs) in different Tx/Rx combinations. When I further decrease the used bandwidth/sample rate, I get the a similar transfer function but only the corresponding portion of the one at 10 MHz band width - in the picture the share from -2.5 MHz to +2.5 MHz at same magnitude. Thus the problem is not in the signal processing done in GNU radio, but must be either in the DSP on FPGA/N210 or in the RF path of the SBX. I cannot imagine that such a large frequency selectivity in DSP/RF hardware is wanted. The SBX is specified for 40 MHz bandwith. And I doubt that this is a sinc D/A-C distortion, because 10 MHz is a tenth of the D/A-C sample rate.
Could you give me some hint for an explanation, please? Any help in changing this?
Best regards
Stephan Ludwig
Communication Technology (CR/AEH4)
Robert Bosch GmbH | Renningen | 70465 Stuttgart | GERMANY |
www.bosch.comhttp://www.bosch.com Tel. +49(711)811-8809tel:%2B49%28711%29811-8809 | Mobile +49(172)5630639tel:%2B49%28172%295630639 | Fax
+49(711)811-5187845tel:%2B49%28711%29811-5187845 | Stephan.Ludwig2@de.bosch.commailto:Stephan.Ludwig2@de.bosch.com
Registered Office: Stuttgart, Registration Court: Amtsgericht
Stuttgart, HRB 14000; Chairman of the Supervisory Board: Franz
Fehrenbach; Managing Directors: Dr. Volkmar Denner, Dr. Stefan
Asenkerschbaumer, Dr. Rolf Bulander, Dr. Stefan Hartung, Dr. Markus
Heyn, Dr. Dirk Hoheisel, Christoph Kübel, Uwe Raschke, Dr. Werner
Struth, Peter Tyroller
<CableTransfer.png>_______________________________________________
USRP-users mailing list
USRP-users@lists.ettus.commailto:USRP-users@lists.ettus.com
http://lists.ettus.com/mailman/listinfo/usrp-users_lists.ettus.com