Hi Andrey,

On Mon, Oct 29, 2012 at 6:46 AM, Andrey Sviyazov andreysviyaz@gmail.com wrote:

Do you use TXLPF and which if yes? And also, please let us know other registers dump of TX part, especially TXVGA's. Btw, unfortunatelly, I do not know how to switch LPF otherwise than by direct registers access. Now I work with power consumption and forced to write to registers to see can LPF to suppress spurs (which occur without LDO regulators) or not.

I set LPF through the UHD interface. This was necessary because the default setting was too narrow. Unfortunately, I did not capture at any register values.

https://github.com/ttsou/openbts-multi-arfcn/commit/a95d8f7f

I don't see how the LPF can suppress these spurs because they occur inside the multicarrier passband.

Thomas

Hi Alexander,

It is IQ imbalance indeed. It has to be corrected in BB. I believe you have access to the document:

"LMS6002D Improving transceiver performance using digital techniques -1.0r1.pdf"

It may be useful. Also, we recommend to implement RX DC auto cancellation as explained in section 3.3 of the same document. This should solve RX LO leakage issue.

We usually measure < -60dBc unwanted image level when IQ imbalance is corrected.

Regards, Srdjan

On 29/10/2012 12:33, Alexander Chemeris wrote:

Hi Srdjan,

Have you seen a question at the end of this e-mail, regarding the LMS calibration? To me it looks like a result of the I/Q imbalance which we should compensate at the baseband level. Is that correct?

On Thu, Oct 25, 2012 at 10:16 PM, Thomas Tsou ttsou@vt.edu wrote:

...

Hi all,

I spent some time yesterday testing UmTRX with MCBTS. Aside from some trivial fixes (inverted sample rate), the code ran fine with no issues. The default bandwidth of the LPF was insufficient, so that was widened. I tested with 9 handsets between 3 and 7 carriers on a Core2 Duo (P7570) laptop.

git://github.com/ttsou/openbts-multi-arfcn.git umtrx

Configuration is through sqlite database, which is unchanged from mainline; just set the number of ARFCN (7 max) and C1 channels. Because of the bandwidth involved, be sure to increase the maximum buffer sizes for UHD. The uhd_usrp_probe utility will warn if the target size cannot be set.

http://files.ettus.com/uhd_docs/manual/html/transport.html#linux-specific-no...

Design

There are three resampling stages.

1. Outer resampler converts from the device rate to a multiple of the

400 kHz channel spacing. 2. Channelizer demultiplexes input stream to M channels 3. Inner resampler converts from 400 kHz to a multiple of the GSM symbol rate.

UmTRX rates:

Channelizer Rate Device Rate Decimation

1.6 Msps            1.625 Msps         8
3.2 Msps            3.250 Msps         4

Options

To broadcast dummy bursts on all carriers for spectrum testing, uncomment the following preprocessor declaration in radioParams.h.

#define ENABLE_ALL_CHANS

With dummy bursts on all carriers:

http://filebox.vt.edu/users/ttsou/http/7carrier_umtrx.PNG

Calibration

Calibration remains a concern. With single carrier, the carrier leakage and and IQ imbalance appear as in-band distortion. For MCBTS, OpenBTS offsets C0 on to the lowest carrier. This makes the carrier and image quite visible, as shown in the following uncalibrated capture. I checked with a swept tone to verify that this issue was not a result of baseband DSP. The USRP also shows similar effects.

http://filebox.vt.edu/users/ttsou/http/carrier_image.PNG

Any idea how much carrier and image suppression we should expect with calibration on UmTRXv2?

Thomas

I am not sure calibration precision is affected by sample rate. Theoretically, we are rotating I vector by alpha/2 and Q vector by -alpha/2 and that is done by multiplication. Hence, the resolution of quadrature phase error (alpha) which we can correct is defined by the with of multipliers.

Any way, we clocked ADCs at 15.36MHz (30.72MS/s digital IQ interface) in the experiment I mentioned.

On 29/10/2012 13:26, Alexander Chemeris wrote:

Hi Srdjan,

If I understand correctly, I/Q imbalance calibration precision depends on the sampling rate. What sampling rate do you use to achieve <-60dBc image level?

On Mon, Oct 29, 2012 at 5:19 PM, Srdjan Milenkovic s.milenkovic@limemicro.com wrote:

...

Hi Alexander,

It is IQ imbalance indeed. It has to be corrected in BB. I believe you have access to the document:

"LMS6002D Improving transceiver performance using digital techniques -1.0r1.pdf"

It may be useful. Also, we recommend to implement RX DC auto cancellation as explained in section 3.3 of the same document. This should solve RX LO leakage issue.

We usually measure < -60dBc unwanted image level when IQ imbalance is corrected.

Regards, Srdjan

On 29/10/2012 12:33, Alexander Chemeris wrote:

...

Hi Srdjan,

Have you seen a question at the end of this e-mail, regarding the LMS calibration? To me it looks like a result of the I/Q imbalance which we should compensate at the baseband level. Is that correct?

On Thu, Oct 25, 2012 at 10:16 PM, Thomas Tsou ttsou@vt.edu wrote:

...

Hi all,

I spent some time yesterday testing UmTRX with MCBTS. Aside from some trivial fixes (inverted sample rate), the code ran fine with no issues. The default bandwidth of the LPF was insufficient, so that was widened. I tested with 9 handsets between 3 and 7 carriers on a Core2 Duo (P7570) laptop.

git://github.com/ttsou/openbts-multi-arfcn.git umtrx

Configuration is through sqlite database, which is unchanged from mainline; just set the number of ARFCN (7 max) and C1 channels. Because of the bandwidth involved, be sure to increase the maximum buffer sizes for UHD. The uhd_usrp_probe utility will warn if the target size cannot be set.

http://files.ettus.com/uhd_docs/manual/html/transport.html#linux-specific-no...

Design

There are three resampling stages.

1. Outer resampler converts from the device rate to a multiple of the

400 kHz channel spacing. 2. Channelizer demultiplexes input stream to M channels 3. Inner resampler converts from 400 kHz to a multiple of the GSM symbol rate.

UmTRX rates:

Channelizer Rate Device Rate Decimation

 1.6 Msps            1.625 Msps         8
 3.2 Msps            3.250 Msps         4

Options

To broadcast dummy bursts on all carriers for spectrum testing, uncomment the following preprocessor declaration in radioParams.h.

#define ENABLE_ALL_CHANS

With dummy bursts on all carriers:

http://filebox.vt.edu/users/ttsou/http/7carrier_umtrx.PNG

Calibration

Calibration remains a concern. With single carrier, the carrier leakage and and IQ imbalance appear as in-band distortion. For MCBTS, OpenBTS offsets C0 on to the lowest carrier. This makes the carrier and image quite visible, as shown in the following uncalibrated capture. I checked with a swept tone to verify that this issue was not a result of baseband DSP. The USRP also shows similar effects.

http://filebox.vt.edu/users/ttsou/http/carrier_image.PNG

Any idea how much carrier and image suppression we should expect with calibration on UmTRXv2?

Thomas

Forgot to mention, < -60dBc image suppression has been achieved by correcting both phase and gain imbalance.

On 30/10/2012 10:10, Alexander Chemeris wrote:

Thank you. We will check how well we could compensate I/Q imbalance on UmTRX.

On Mon, Oct 29, 2012 at 5:40 PM, Srdjan Milenkovic s.milenkovic@limemicro.com wrote:

...

I am not sure calibration precision is affected by sample rate. Theoretically, we are rotating I vector by alpha/2 and Q vector by -alpha/2 and that is done by multiplication. Hence, the resolution of quadrature phase error (alpha) which we can correct is defined by the with of multipliers.

Any way, we clocked ADCs at 15.36MHz (30.72MS/s digital IQ interface) in the experiment I mentioned.

On 29/10/2012 13:26, Alexander Chemeris wrote:

...

Hi Srdjan,

If I understand correctly, I/Q imbalance calibration precision depends on the sampling rate. What sampling rate do you use to achieve <-60dBc image level?

On Mon, Oct 29, 2012 at 5:19 PM, Srdjan Milenkovic s.milenkovic@limemicro.com wrote:

...

Hi Alexander,

It is IQ imbalance indeed. It has to be corrected in BB. I believe you have access to the document:

"LMS6002D Improving transceiver performance using digital techniques -1.0r1.pdf"

It may be useful. Also, we recommend to implement RX DC auto cancellation as explained in section 3.3 of the same document. This should solve RX LO leakage issue.

We usually measure < -60dBc unwanted image level when IQ imbalance is corrected.

Regards, Srdjan

On 29/10/2012 12:33, Alexander Chemeris wrote:

...

Hi Srdjan,

Have you seen a question at the end of this e-mail, regarding the LMS calibration? To me it looks like a result of the I/Q imbalance which we should compensate at the baseband level. Is that correct?

On Thu, Oct 25, 2012 at 10:16 PM, Thomas Tsou ttsou@vt.edu wrote:

...

Hi all,

I spent some time yesterday testing UmTRX with MCBTS. Aside from some trivial fixes (inverted sample rate), the code ran fine with no issues. The default bandwidth of the LPF was insufficient, so that was widened. I tested with 9 handsets between 3 and 7 carriers on a Core2 Duo (P7570) laptop.

 git://github.com/ttsou/openbts-multi-arfcn.git umtrx

Configuration is through sqlite database, which is unchanged from mainline; just set the number of ARFCN (7 max) and C1 channels. Because of the bandwidth involved, be sure to increase the maximum buffer sizes for UHD. The uhd_usrp_probe utility will warn if the target size cannot be set.

http://files.ettus.com/uhd_docs/manual/html/transport.html#linux-specific-no...

Design

There are three resampling stages.

1. Outer resampler converts from the device rate to a multiple of the

400 kHz channel spacing. 2. Channelizer demultiplexes input stream to M channels 3. Inner resampler converts from 400 kHz to a multiple of the GSM symbol rate.

UmTRX rates:

Channelizer Rate Device Rate Decimation

  1.6 Msps            1.625 Msps         8
  3.2 Msps            3.250 Msps         4

Options

To broadcast dummy bursts on all carriers for spectrum testing, uncomment the following preprocessor declaration in radioParams.h.

#define ENABLE_ALL_CHANS

With dummy bursts on all carriers:

 http://filebox.vt.edu/users/ttsou/http/7carrier_umtrx.PNG

Calibration

Calibration remains a concern. With single carrier, the carrier leakage and and IQ imbalance appear as in-band distortion. For MCBTS, OpenBTS offsets C0 on to the lowest carrier. This makes the carrier and image quite visible, as shown in the following uncalibrated capture. I checked with a swept tone to verify that this issue was not a result of baseband DSP. The USRP also shows similar effects.

 http://filebox.vt.edu/users/ttsou/http/carrier_image.PNG

Any idea how much carrier and image suppression we should expect with calibration on UmTRXv2?

 Thomas