usrp-users@lists.ettus.com
Discussion and technical support related to USRP, UHD, RFNoC
View all threadsChoice of central frequencies
We have our LTE stack up and running, both UE and eNB on B210 boards. The
peak rate performance (achieved when the boards are fairly close) depends
too much on the central frequencies selected for uplink/downlink
transmissions. We know from other measurements that all frequencies tested
are interference-free and see roughly the same noise floor, hence the
problem may come from the boards themselves. Nothing in the code changes
across runs except for the central frequencies.
Given a master clock rate (for example, 15.36 MHz) and a sampling rate (for
example, 1.92 MHz = 15.36 / 8 MHz), is there a B210-friendly rule that
allows me to pick a pair of frequencies that will work as uplink/downlink
pair for full-duplex use? Something like avoiding integer multiples of the
master clock rate?
Thanks,
Dario
Hi Dario,
On Thu, Apr 30, 2015 at 3:37 PM, Dario Fertonani via USRP-users
usrp-users@lists.ettus.com wrote:
We have our LTE stack up and running, both UE and eNB on B210 boards. The
peak rate performance (achieved when the boards are fairly close) depends
too much on the central frequencies selected for uplink/downlink
transmissions. We know from other measurements that all frequencies tested
are interference-free and see roughly the same noise floor, hence the
problem may come from the boards themselves. Nothing in the code changes
across runs except for the central frequencies.
What are some of the better/worse frequencies that you have tested?
And can you extract any details from the stack indicating why peak
performance numbers are dropping (e.g. dropped frames, low MCS values,
synchronization failures).
If you are relying on the internal clock reference, reduced
performance at upper frequencies rather than lower frequencies would
make suspect frequency offset and drift. Also, RF gain settings (and
antennas if you are using them) are dependent on frequency, which may
be a factor.
Given a master clock rate (for example, 15.36 MHz) and a sampling rate (for
example, 1.92 MHz = 15.36 / 8 MHz), is there a B210-friendly rule that
allows me to pick a pair of frequencies that will work as uplink/downlink
pair for full-duplex use? Something like avoiding integer multiples of the
master clock rate?
The RF frequency is handled by the AD9361 and FPGA cordic, which are
not affected by decimation/interpolation rates.
-TT
Hi Tom,
A short summary first. We use the tuning API with LO forced to half of the
sampling rate, to avoid signal degradation around DC. I can't understand
why it matters so much what carrier frequency we pick, all within an
"empty" spectrum. I was wondering if this is expected and if we have to
restrict ourselves to frequencies that satisfy some equation involving the
master clock rate, or the signal bandwidth, or the like.
Please find my answers below. The behavior I'm going to describe is
perfectly repeatable.
We pick two frequency ranges known to be "empty" enough. Eventually we find
two frequencies on which our LTE stack gets peak-rate both in uplink and
downlink, in a full-duplex MIMO setting. This is expected and desired when
the devices sit next to each other. However, I can't explain the following.
Weird behavior #1. If we switch UL and DL frequencies, we lose peak rate in
both directions. MCS and SNR drop, etc. This rules out the "we picked a bad
carrier" hypothesis, as the two devices are identical.
Weird behavior #2. If we shift any carrier frequency by even just 1 MHz
(still within the "empty" spectrum) the MCS and SNR may drop. Then we shift
by another MHz and the MCS and SNR may go back up.
It seems to be a narrowband issue, as this effect tends to be less
noticeable if we use wider band LTE signals (e.g., this behavior is
dramatic in 1.25 MHz LTE but much less in 5 MHz LTE). Because of how LTE
channel coding works, this behavior is compatible with a signal that has
poor quality on just a few OFDM carriers (some tens of KHz).
Thanks,
Dario
On Mon, May 18, 2015 at 12:09 PM, Tom Tsou tom.tsou@ettus.com wrote:
Hi Dario,
On Thu, Apr 30, 2015 at 3:37 PM, Dario Fertonani via USRP-users
usrp-users@lists.ettus.com wrote:
We have our LTE stack up and running, both UE and eNB on B210 boards. The
peak rate performance (achieved when the boards are fairly close) depends
too much on the central frequencies selected for uplink/downlink
transmissions. We know from other measurements that all frequencies
tested
are interference-free and see roughly the same noise floor, hence the
problem may come from the boards themselves. Nothing in the code changes
across runs except for the central frequencies.
What are some of the better/worse frequencies that you have tested?
And can you extract any details from the stack indicating why peak
performance numbers are dropping (e.g. dropped frames, low MCS values,
synchronization failures).
If you are relying on the internal clock reference, reduced
performance at upper frequencies rather than lower frequencies would
make suspect frequency offset and drift. Also, RF gain settings (and
antennas if you are using them) are dependent on frequency, which may
be a factor.
Given a master clock rate (for example, 15.36 MHz) and a sampling rate
(for
example, 1.92 MHz = 15.36 / 8 MHz), is there a B210-friendly rule that
allows me to pick a pair of frequencies that will work as uplink/downlink
pair for full-duplex use? Something like avoiding integer multiples of
the
master clock rate?
The RF frequency is handled by the AD9361 and FPGA cordic, which are
not affected by decimation/interpolation rates.
-TT
Dario,
I strongly suggest tuning the LO to the center of your LTE signal.
LTE was designed to handle this, and it is the recommended mode of
operation of the chip. It is how we use it, and it has supported all
of the MCS'es quite well.
Matt
On Tue, May 19, 2015 at 9:34 AM, Dario Fertonani via USRP-users
usrp-users@lists.ettus.com wrote:
Hi Tom,
A short summary first. We use the tuning API with LO forced to half of the
sampling rate, to avoid signal degradation around DC. I can't understand why
it matters so much what carrier frequency we pick, all within an "empty"
spectrum. I was wondering if this is expected and if we have to restrict
ourselves to frequencies that satisfy some equation involving the master
clock rate, or the signal bandwidth, or the like.
Please find my answers below. The behavior I'm going to describe is
perfectly repeatable.
We pick two frequency ranges known to be "empty" enough. Eventually we find
two frequencies on which our LTE stack gets peak-rate both in uplink and
downlink, in a full-duplex MIMO setting. This is expected and desired when
the devices sit next to each other. However, I can't explain the following.
Weird behavior #1. If we switch UL and DL frequencies, we lose peak rate in
both directions. MCS and SNR drop, etc. This rules out the "we picked a bad
carrier" hypothesis, as the two devices are identical.
Weird behavior #2. If we shift any carrier frequency by even just 1 MHz
(still within the "empty" spectrum) the MCS and SNR may drop. Then we shift
by another MHz and the MCS and SNR may go back up.
It seems to be a narrowband issue, as this effect tends to be less
noticeable if we use wider band LTE signals (e.g., this behavior is dramatic
in 1.25 MHz LTE but much less in 5 MHz LTE). Because of how LTE channel
coding works, this behavior is compatible with a signal that has poor
quality on just a few OFDM carriers (some tens of KHz).
Thanks,
Dario
On Mon, May 18, 2015 at 12:09 PM, Tom Tsou tom.tsou@ettus.com wrote:
Hi Dario,
On Thu, Apr 30, 2015 at 3:37 PM, Dario Fertonani via USRP-users
usrp-users@lists.ettus.com wrote:
We have our LTE stack up and running, both UE and eNB on B210 boards.
The
peak rate performance (achieved when the boards are fairly close)
depends
too much on the central frequencies selected for uplink/downlink
transmissions. We know from other measurements that all frequencies
tested
are interference-free and see roughly the same noise floor, hence the
problem may come from the boards themselves. Nothing in the code changes
across runs except for the central frequencies.
What are some of the better/worse frequencies that you have tested?
And can you extract any details from the stack indicating why peak
performance numbers are dropping (e.g. dropped frames, low MCS values,
synchronization failures).
If you are relying on the internal clock reference, reduced
performance at upper frequencies rather than lower frequencies would
make suspect frequency offset and drift. Also, RF gain settings (and
antennas if you are using them) are dependent on frequency, which may
be a factor.
Given a master clock rate (for example, 15.36 MHz) and a sampling rate
(for
example, 1.92 MHz = 15.36 / 8 MHz), is there a B210-friendly rule that
allows me to pick a pair of frequencies that will work as
uplink/downlink
pair for full-duplex use? Something like avoiding integer multiples of
the
master clock rate?
The RF frequency is handled by the AD9361 and FPGA cordic, which are
not affected by decimation/interpolation rates.
-TT
USRP-users mailing list
USRP-users@lists.ettus.com
http://lists.ettus.com/mailman/listinfo/usrp-users_lists.ettus.com
On Tue, May 19, 2015 at 9:34 AM, Dario Fertonani
dario.fertonani@gmail.com wrote:
A short summary first. We use the tuning API with LO forced to half of the
sampling rate, to avoid signal degradation around DC.
DC offset shouldn't be an issue. The LTE downlink doesn't load the
center OFDM carrier and the uplink has a half carrier spacing shift.
The AD9361 compensates well in this environment as well.
It seems to be a narrowband issue, as this effect tends to be less
noticeable if we use wider band LTE signals (e.g., this behavior is dramatic
in 1.25 MHz LTE but much less in 5 MHz LTE). Because of how LTE channel
coding works, this behavior is compatible with a signal that has poor
quality on just a few OFDM carriers (some tens of KHz).
Since you're performing end-to-end testing, I'm guessing that access
to physical layer status - mainly the PDSCH constellation and channel
equalizer estimate - is not straightforward. That would be immensely
helpful though as narrowband interference or distortion will show up
clearly in the latter.
Do you have access to spectrum analyzer or another USRP to act as one?
-TT
Tom, Matt,
Thank you for your answers.
Yes, I have access to every PHY info, since our entire stack is in
software. I know that LTE is robust to DC component problems, but we had to
move to non-zero LO precisely because a few OFDM carriers around the DC one
gave terrible SNR and soft symbol constellation. That did improve things.
All of this is for eNB and UE each running on x86 with a B210 connected to
it. The master clock rate is set to 15.36 MHz, although I read on this list
that the bug affecting full-duplex MIMO at master clock rate 30.72 MHz is
fixed, so we'll eventually move to that value. We run the B210 sometime on
USB power and sometime on external power - I know the former is somewhat
out of spec for full-duplex MIMO, but nothing seems to change unless we go
really high on carrier freqs and sampling rates.
Anyway, we have now frozen the RF part of the code (UHD_003.008.003 master,
plus our wrappers), since we found a frequency setting that's very good.
I'll open a new thread later, if the problem persists with a newer UHD,
after unfreezing. For now it's sufficient (and very good!) for us to know
that the extreme sensitivity of the signal quality to the carrier frequency
within an empty spectrum is not a limitation of the board, but a bug
somewhere in the code (UHD, or ours).
Best,
Dario
On Tue, May 19, 2015 at 11:40 AM, Tom Tsou tom.tsou@ettus.com wrote:
On Tue, May 19, 2015 at 9:34 AM, Dario Fertonani
dario.fertonani@gmail.com wrote:
A short summary first. We use the tuning API with LO forced to half of
the
sampling rate, to avoid signal degradation around DC.
DC offset shouldn't be an issue. The LTE downlink doesn't load the
center OFDM carrier and the uplink has a half carrier spacing shift.
The AD9361 compensates well in this environment as well.
It seems to be a narrowband issue, as this effect tends to be less
noticeable if we use wider band LTE signals (e.g., this behavior is
dramatic
in 1.25 MHz LTE but much less in 5 MHz LTE). Because of how LTE channel
coding works, this behavior is compatible with a signal that has poor
quality on just a few OFDM carriers (some tens of KHz).
Since you're performing end-to-end testing, I'm guessing that access
to physical layer status - mainly the PDSCH constellation and channel
equalizer estimate - is not straightforward. That would be immensely
helpful though as narrowband interference or distortion will show up
clearly in the latter.
Do you have access to spectrum analyzer or another USRP to act as one?
-TT