|Subject:||Re: [Discuss-gnuradio] Feedback with Transmitters and Receiver|
|Date:||Tue, 3 May 2016 18:06:10 -0700|
Hi Derek,Thanks for getting back to me. I lowered the Tx gain by about 15 dB on each channel, and then now it's looking good. It makes sense that it was clipping. Using your flowgraph (thanks!), I am appearing to get good results for setup (2), which is what we want. Looking at phase_error.png, we are getting a phase error of about -60 dB, which is very good. And we can see from i_and_q_both_channels.png that the phase offset is about 60 degrees (pi/3) (looking at both I and Q). I think this shows that both transmitters are transmitting at the same time. If I applied accurate beamforming weights, then ideally that phase offset would be eliminated (which is what I plan to do). I just want to make sure from you that this behavior makes sense.
I believe my next step would be to look into timed commands in UHD and tags in GNU Radio to see how I can now transmit something from these two aligned transmitters to just one receiver, then calculate that power at the receiver, and then send it back to the transmitters for some processing before transmitting again.--On Tue, May 3, 2016 at 4:05 PM, Derek Kozel <address@hidden> wrote:DerekCheers,Test 2, where the receivers are synchronized with 1PPS and 10 MHz references from the same octoclock as the transmitters (or another GPSDO of similar grade) will give you time, frequency, and phase (with SBXs) information.Test 1, where the receivers are not synchronized, will tell you none of time, frequency, or phase alignment since the receivers could be in any state compared to each other and the synchronized transmitter pair.Hello Pavan,Your signal strength is much too high. You can see on the plots that the received signal is greater than 1 (assuming a sinusoid input). Lower the transmit gain and/or the receive gain until you are no longer clipping. The ADC is being overloaded so the spikes are entirely to be expected. You have the 20dB attenuation which should be preventing any damage, but the N210+SBX is still plenty sensitive enough to clip.I suggest not using complex to float conversion blocks and looking at the raw complex values. It should be easy to see time and frequency sink with the time sink.To measure phase alignment you can take the two outputs of your source block and use a conjugate multiply to view the relative phase offset. I've attached a flowgraph which I use to prove B210 phase alignment. You should be able to copy the flowgraph over to your setup pretty simply.On Tue, May 3, 2016 at 3:51 PM, Pavan Yedavalli <address@hidden> wrote:From (1), which I think is the right test to confirm that the transmitters are transmitting at the same time, we see the constant phase offset that we expect. From (2), things get a bit weird and I'm not sure how to interpret those results.2.) Both receivers are connected to the Octoclock (separately directly to the Octoclock, not one connected to the other via MIMO cable and only one connected to Octoclock), and all of the time/sync properties and PPS are what you had described in a previous e-mail (see received_connected_to_octoclock.png). Looking at recvd_real_parts_octo_on_recv.png and recvd_real_parts_octo_on_recv_2.png, we can see that the frequencies appear to be slightly different (I think?), and that there is also a spike in the channel 2 capture as well, so I'm not sure what is happening while the Octoclock is connected on the receive side.1.) Neither receiver is connected to the Octoclock, and instead all of the time/sync properties are set to Default and PPS is set to don't sync (see received_not_connected_to_octoclock.png). We can see from recv_real_parts_no_octo_on_recv.png that they seem to be offset in phase slightly, which is what we expect, but the frequencies appear the same. One issue I'm seeing is that there are these periodic spikes in the channel 1 curve - is that because it may not be connected perfectly? I'm not sure where those are coming from.Hi Derek,I was able to put in another USRP as a receiver, so I have the same transmit setup (connected to Octoclock and with MIMO cable to another USRP), and the receive setup is now 2 USRPs, not connected by anything, and in two different setups:Please keep me posted if you have any insight regarding (a) the periodic spikes, (b) whether (1) confirms what we want it to, and (c) whether I should be concerned about (2), since intuitively it makes sense that that should behave the same as (1), right? Thank you again for all your help.--On Fri, Apr 29, 2016 at 5:37 PM, Derek Kozel <address@hidden> wrote:Perhaps capturing the combined waveform at one frequency, tuning only the transmitters to a different frequency and then back to the original and comparing the waveforms would be sufficient? The phase alignment will be a constant offset at a given frequency, assuming nothing else in the system changes.Hi Pavan,Yes, you are correct, this test won't show phase alignment. The most common checks for phase alignment need a receiver able to do two simultaneous channels, such as a high bandwidth oscilloscope, another N210/SBX that can be setup the same as your transmitter pairing, or an X300 series USRP with two SBXs. Perhaps others on the list will have less equipment ways of verifying phase alignment.On Fri, Apr 29, 2016 at 5:14 PM, Pavan Yedavalli <address@hidden> wrote:Hi Derek,That makes sense. I will put a combiner and try this. However, now the test is a bit different, right? The only way I could tell that the transmitters are transmitting at the same time is if the power level is double what it used to be, assuming they are actually fully in phase. Is there a possibility that they are out of phase though, but there is a constant random offset, so they add up slightly differently and not completely coherently? (as shown in Figure 7 in this document)Thanks again.--On Fri, Apr 29, 2016 at 12:21 PM, Derek Kozel <address@hidden> wrote:DerekYou will need to use a combiner to merge the two transmit signals into a single receive connection, to either TX/RX or RX2. Or use antennas and let the air do your combining. You may want to keep the attenuators in line until you are comfortable with the power levels. You'll be able to see when your receiver is clipping in the Scope GUI.Hi Pavan,I'm sorry, I didn't read your cabling closely enough or I would have noticed this before. If you only have one SBX in the receive N210, then you only have one possible receive channel! That is why you are seeing the error that RX channel 1 (remember they're 0 indexed) is out of range. On the SBX, and all other daughterboards, the TX/RX and RX2 ports are shared by a single receive channel. They are setup with a switch to allow either time divided transmit and receive on a single antenna (using the TX/RX port) or having separate transmit and receive connections (transmit on TX/RX and receive on RX2).On Fri, Apr 29, 2016 at 11:32 AM, Pavan Yedavalli <address@hidden> wrote:Hi Derek,I made all the changes, and the Tx error as well as the warnings are now gone. However, the Rx error still remains:File "/usr/local/lib/python2.7/dist-packages/gnuradio/uhd/uhd_swig.py", line 2168, in set_samp_rate
return _uhd_swig.usrp_source_sptr_set_samp_rate(self, *args, **kwargs)RuntimeError: LookupError: IndexError: multi_usrp: RX channel 1 out of range for configured RX frontendsI'm not entirely sure what the problem is here. Attached are the flowgraph pictures. In addition, I am using Rev 5.1 SBX daughterboards. Thanks again.--On Thu, Apr 28, 2016 at 4:21 PM, Pavan Yedavalli <address@hidden> wrote:Hi Derek,I appreciate it. Okay, I will change all of those. I am using the SBX daughterboards - the ones that support phase sync.--On Thu, Apr 28, 2016 at 3:55 PM, Derek Kozel <address@hidden> wrote:DerekRegards,Hello Pavan,You do not need Ethernet connected to the Octoclock, so that's ok. Your cabling sounds correct. Can you please combine both UHD Sinks? Just increase the number of motherboards and channels to 2 and copy the MIMO attached USRP's settings into channel 2.What daughterboards are you using? On the N200 series motherboards only the SBX daughterboards supports phase synchronization. What you should see is frequency and time synchronization between the MIMO N210s.
Your sample rate for the receive side is very very low. I suspect that will throw a warning if you read the log output at the bottom of GRC. Try raising that to 500kHz or more. Also the WX Scope Sink can be changed to complex inputs so you don't need the converter blocks. You are also setting a custom clock rate. The N200 has a fixed master clock rate of 100 MHz so this is likely also throwing a warning in the log messages. Definitely look through those and make changes as needed.On Thu, Apr 28, 2016 at 12:22 PM, Pavan Yedavalli <address@hidden> wrote:Hi Derek,Sorry - just another quick addition. When I run the Tx flowgraph, I get this error:Board 0 May not be getting a PPS signal! No PPS detected within the time interval.This definitely tells me something is wrong with the Octoclock-G setup.--On Thu, Apr 28, 2016 at 12:18 PM, Pavan Yedavalli <address@hidden> wrote:The setup:Hi Derek,I am trying to do (3), as you noted above, and my test to see whether the Tx USRPs are transmitting at the same time is to directly connect them to the Rx USRP and plot the real components of each one and see whether they are in phase (or at least with some constant random offset). In theory, I believe this is a good test to see that the Octoclock-G is working its magic.Octoclock: Two of the three boards (Master Tx USRP and 1 Rx USRP) are connected to the Octoclock-G (one cable each from PPS out to PPS in, and one cable each from 10 MHz out to Ref in, so 4 total cables). The primary ref knob is set to Internal, and the PPS blinks green, while Internal, Status, and Power are all steady greens. I do not have the ethernet of the Octoclock connected, however. When I connected it to my Gb Ethernet switch, the indicator was orange, while all the other working ones are green, so I decided not to connect it for now. Does this matter?Tx side: I have both Tx USRPs connected to each other via the MIMO cable, and one of them is connected to the Octoclock, as mentioned above. In tx_mimo_setup_octoclock.png, we can see that I have two USRP sinks connected via MIMO cable, and one of them has time and clock set to External, and the other has time and clock set to MIMO cable. Both have sync set to Unknown PPS.Between Tx and Rx: I have an SMA cable from one Tx USRP connected to a 20 dB attenuator, and then connected to the Tx/Rx port of the Rx USRP. I have another SMA cable from the other Tx USRP connected to a 20 dB attenuator, and then connected to the Rx2 port of the Rx USRP. No antennas are connected.
Rx side: In receiver_recvng_on_both_ports.png, we can see that I have a USRP source with two channels. The time and clock are set to External, and the sync is set to Unknown PPS. I run this, but I get the following error:I tried looking up what this error is, and apparently there was a fix in a branch years ago, but I'm assuming I have that fix already? I have a feeling something is wrong with the Octoclock setup that is causing this, but I'm not sure. I believe the setup I mentioned above makes sense, right?
RuntimeError: LookupError: IndexError: multi_usrp: RX channel 1 out of range for configured RX frontendsObviously, I will look into timed commands in UHD and tags in GNU Radio after I get all of this set up and working. Thank you so much again for the help.--On Thu, Apr 21, 2016 at 3:52 PM, Derek Kozel <address@hidden> wrote:DerekRegards,Once you've given this a try let us know if you have additional questions.If this is your first use of USRPs and GNU Radio then I'd suggest reading through the tutorials available online and not get too focused on MIMO until you feel comfortable with the basics of the environment and tools that you have.Here's a page that talks about synchronization of USRPs. Read this, get your hardware all setup, and try setting up a basic GRC flowgraph with your three radios. Think of what tests you could use to verify that both your MIMO cabled radios are transmitting at the same time. You should look into timed commands in UHD and tags in GNU Radio.3) Once you connect all the Octoclock signals then in GNU Radio you can select the Clock and Time sources to be External and the Sync to be Unknown PPS. Your pair of units connected via a MIMO cable are special, the master should have the External time and clock sources, the companion USRP should have MIMO selected for time and clock. The Sync should still be Unknown PPS.Hi Pavan,
This is a USRP/UHD question really so I'm including the usrp-users mailing list. If you're not already the list already then you should certainly join as that's a better resource for questions about UHD/USRPs.2) Most radio systems don't have GPS timebases available and use various protocol level methods for aligning their clocks, if needed. In a very simple system the receiver could simply listen continuously until it receives a full message, then transmits a response if needed. Look up Time Division Multiplexing and Frequency Division Multiplexing. This is an area where there are nearly as many possibilities as there are radio systems.
1) Any SMA cable will work. For the best performance their electrical lengths should be the same. In practice this usually means equal physical lengths of the same type of coax. This ensures that the signals arrive at the same time (and phase).
http://gnuradio.org/redmine/projects/gnuradio/wiki/Guided_TutorialsOn Thu, Apr 21, 2016 at 3:27 PM, Pavan Yedavalli <address@hidden> wrote:Hi Derek,Thanks for getting back to me. So, I do have an Octoclock, so I think we're getting somewhere and this is starting to make more sense. A few follow up questions:1.) Do I need special cables to connect all of the units to the Octoclock, or are they robust SMA cables?2.) I feel like this seems particularly involved to send a signal from a transmitter to a receiver. I am assuming most non-MIMO, non-beamforming related tasks have always used your second option of using the GPSDO kits? I purchased an Octoclock knowing I would do MIMO experiments, but obviously I'm guessing more conventional communication techniques (like a simple BPSK or QPSK between tx and rx) would have probably used the GPSDO kits?3.) Once I connect them all to the Octoclock, then I don't need to a protocol level time synchronization, right? Once they're all synchronized and I see that in the plots, then I guess the next step would be to figure out how to implement my actual feedback loop. At that point, then I would need to figure out how to do burst mode to transmit and receiver timed signals? Would this end up needing to be one flow graph or would I have to use two flow graphs? (One for to and one for rx, the way I am doing it now)Thank you again for all the help. I think I'm starting to understand what I need in the setup.--On Thu, Apr 21, 2016 at 4:56 PM, Derek Kozel <address@hidden> wrote:DerekRegards,There are many ways of implementing a protocol level time synchronization in software/DSP. The paper I linked to talks about one way, there are certainly others. I do not know of any example projects implementing them though so you would have to develop your own.Connect GPS referenced 1PPS signals to both the master USRP of your MIMO cabled pair and the receiver (For example using two of the GPSDO kits: https://www.ettus.com/product/details/GPSDO-KIT)ORConnect a shared 1PPS signal to *both* the master USRP of your MIMO cabled pair and to the receiver (For example using an octoclock: https://www.ettus.com/product/details/OctoClock-G)Hello Pavan,I think we both are starting to understand the setup and the problem. Here are the two hardware solutions:On Thu, Apr 21, 2016 at 8:04 AM, Pavan Yedavalli <address@hidden> wrote:Hi Derek,I'll answer your questions in-line, because I think what you are saying is beginning to make me understand what I need:On Wed, Apr 20, 2016 at 9:03 PM, Derek Kozel <address@hidden> wrote:Hello Pavan,Are you trying to create a shared timebase between the two USRPs without having a shared 1PPS or GPS reference? You are still not using enough detail for us to understand fully.To clarify, my setup is two USRPs connected via MIMO cable, and then another USRP acting as a receiver. So are you asking whether I'm trying to create a shared timebase between the two-USRP unit (because they are MIMO cabled) and the receiving USRP without having a shared 1 PPS or GPS reference? I think my answer to that must be yes, because I have not done anything else but connect them to the computer via ethernet and just have two of them connected via MIMO cable and the other one by itself. I'm assuming I need to have a shared reference between the transmit USRPs and the receive USRP, so how would I be able to do that? This could certainly be one of my problems.In Figure 5 both USRPs are connected with a MIMO cable and so have both shared frequency and time bases. What is your weight block doing to the sample stream? Is it a time delay block? I don't know what gnuradio would do if you specified 10*sample_rate as the delay there as that's likely to be a very large number of samples.My weight block is applying a normalized magnitude phase correction to each antenna's transmitted signal, so, yes, it is essentially creating a time delay. Each weight is a complex value with magnitude 1 and a calculated phase. You are saying this could be a problem if it's calculating a value that is too high?If you have both USRPs connected with a time synchronization (shared 1PPS, GPSDO, or MIMO cable) and have your flowgraph configured correctly, then you can just use timed commands to the USRP_alpha to start transmitting at time X and USRP_beta to start receiving at time X and you will see your signal. You can then move to using burst mode using tags to define the number of samples to send/receive along with timed commands to send/receive bursts of samples. This works because the clocks in both USRPs will be aligned to each other.I feel like there are two steps here. First, I need to get the transmitting USRPs (which are conneced via MIMO cable) to time sync to each other (which I believe I have done through using USRP sink in GRC and setting the second channels time and clock to MIMO cable?), and second, I need to get the receive USRP to receive at the same time. So, just as above, I need to get my receive USRP to be on the same time as my transmit USRPs? Once I'm able to do that, then I can do burst mode to transmit and receive timed signals, as you are mentioning?If you do *NOT* have a shared time source for each radio, for instance they are far apart and do not have GPS references, then you need to do some sort of protocol level alignment to create a shared understanding of time between them. A frequently used method is for USRP_alpha to transmit a beacon with a known period (say once every 10 seconds). All other USRPs then receive for longer than 10 seconds to be guaranteed to receive the beacon (assuming they're within range of the transmission). When the receiving USRPs detect the incoming beacon they align their local time to the master (Beacon transmitting) USRP.I guess a similar question to the above: can I have a shared time source between the transmit USRPs (which are already MIMO cabled to each other) and the receive USRP? It seems like that would be easier to do than going through this protocol level alignment, but maybe it's not possible given my setup.Here's a quick paper talking about this topic. The technique is widely used.
http://www.ece.uah.edu/~milenka/docs/dc_ssst05_synch.pdfI hope this helps and is applicable to your need. If you have more questions please try drawing your desired system and maybe include a timeline of events that you expect the radios to do. Attaching your existing flowgraphs, either as photos using GRC's screen capture feature (file>screen capture) or the actual GRC file, also helps us understand what exactly you are working with.I had to take down the setup because I am moving labs, but I will send some flowgraphs and the diagram of the system next week. Thank you again for being so patient and trying to help me. I think I'm just a bit lost on a few of the simple things, but once those are figured out, then I think it should be smoother sailing.DerekOn Wed, Apr 20, 2016 at 4:05 PM, Pavan Yedavalli <address@hidden> wrote:_______________________________________________Hi Martin,I guess I have a few questions:1.) Are there any examples in the gnuradio codebase/flowgraph repository that show how to do synchronized feedback between two USRPs? In other words, I send a signal from a transmit USRP, and then I receive that signal at the receive USRP, and then I send back something else from the receive USRP back to the transmit USRP, and this would be a sequential process in which they are aligned and know when to transmit and/or receive? I saw a post that I think would be relevant, but I'm not sure how to apply it.I believe this should be a pretty standard scenario in which you want to have two USRPs communicate with each other synchronously. I guess I'm just having trouble finding an example of how to do this.2.) Related to the above question, maybe there are no examples to do feedback in one flowgraph, so what I have been doing is the following in my flowgraphs:Flowgraph A:The synchronized MIMO flowgraph (Figure 5) from this, so essentially I have two USRPs synchronized and transmitting out two signals that should be offset but frequency aligned. In my own flowgraph's main(), instead of applying a "phase shift" block, I am applying my own "weights" block to both transmissions.So, I am now sending a signal that has those weights applied to it. So, after I do tb.start(), then I sleep for 10 seconds (by doing sleep(10)) hoping that in the 10 seconds my receiver will catch the signal that I'm transmitting and put it into file.Flowgraph B:My own receiver.py in which I have a USRP sink->FFT->Complex to Mag->File sink. I also have a connection from FFT->QT GUI to see a plot of what is being captured.I now run Flowgraph A in one terminal and Flowgraph B in another terminal. I need to capture A's transmission with the first weights within the 10 seconds (as it's sleeping) into the file sink. Then, A will send a signal with another set of weights applied, and I will need to capture that in the next 10 seconds, and so on. My problem is that I'm often capturing noise because my receive was not aligned with when I was transmitting my desired signal. So, I end up only capturing noise after the transmission stops as opposed to the actual signal when the transmission is happening.Essentially, I am trying to mimic feedback by doing the above, but I don't know how to align my transmitter and receiver, especially because they are two different blocks. Is there a way to make both the transmission and reception one block so that I can do sleep(rx_time + n_samples_since_tag/sampling_rate) (I think this could be right?) as opposed to my static sleep(10) and pray for the best?Would it be helpful at all if I showed you my code? I still feel like I'm not being clear. Sorry about that. If there were any examples, then I think that would be the best for me to look at.Thanks for any help again.--Pavan
Discuss-gnuradio mailing list
|[Prev in Thread]||Current Thread||[Next in Thread]|