IWR6843: Spec and 3D antenna design and BPM/TDM-MIMO Questions

Hi Vincent,

These are some good questions and we'll need some time to get back to you. We should have an answer for you later this week.


Cheers,
Akash

Hi Vincent,

For question 1, there is an error in the documentation for the TIDEP-1000 user's guide. Please use the transmit values and noise figure from the datasheet.

For question 4, BPM is not used in any of the demos. However, the CLI command to enable BPM during configuration exists, and can be found by reading the file at <mmwave_sdk_03_00_00_08>\packages\ti\utils\cli\src\cli_mmwave.c. You can test your BPM configuration with the Out of Box demo.

Regards,

Justin

Hi Justin,

Thanks for reply, how about the question 2 and 3 ?

thanks.

Firstly, I want to clarify that BPM-MIMO is not implemented in 3.0/3.1 release demos. It is only available presently in SDK 2.x on XWR1642 devices (X = A,I) and implements the S1+S2, S1-S2 BPM MIMO scheme and is planned to be implemented in future 3.x release. This method on the SDK 2.x as mentioned in the TI MIMO application note, has a theoretical improvement of 10*log10(2) = 3 dB, what we observed during the demo implementation is a gain between about 2 to 2.5 (the range and noise profile plots of the demo can be observed to see this difference). On the XWR1642 EVM, the two tx-es (there are only two transmit antennas) are separated by 2*lambda. You can read implementation details (what we do in the digital processing chain on the DSP) in the doxygen documentation of the demo.

#2: I am not an antenna expert (and someone who is from TI may respond subsequently if necessary) but the fact that we observed gain in the 2*lambda case as described above tells me that we are not neutralizing gain due to the wider separation. Reading some info regarding this in textbook (Skolnik Intro to radar systems chapter 8), the 2*lambda separation case will produce grating lobes at theta (azimuth angle) = +/-30 deg and +/-90 deg (the 90 one looks much wider and 30 one) for isotropic elements but because the individual elements seems to have a beamwidth of about 60 deg (www.ti.com/.../swru508b.pdf), the grating lobes may be suppressed according to the text and so there may be enough gain at the main lobe compared to the lambda/2 case (where there are no grating lobes)? This is my guess but I may be wrong.

#3: Just to be clear, BPM involves modulating the phase but in the scheme referred, all transmissions are in the same phase i.e phase is not being modulated, only the transmit antennas are enabled/disabled for transmission. This phase modulation is not necessary in this scheme because it is possible to separate the individual transmissions out in the receive digital processing as per the indicated formulas in the picture and thus realize the full 3 Tx entitlement of angular resolution e.g 12 virtual antennas in azimuth) while benefiting from higher transmit power and SNR improvement due to 2 tx simultaneous transmissions. In principle, the full realization of BPM-MIMO for 3 Tx is possible if we can activate all 3 tx-es simultaneously in each chirp and this will require modulation (e.g S1+S2+S3, S1+S2-S3, S1-S2+S3) and similar processing as we do in the 1642 case to separate. In this case, the array separation will be lambda so you don't have the case of 2*lambda. However, 3 tx simultaneous is not supported presently on all devices, as per www.ti.com/.../awr1243.pdf (footnote of table 3-1):
"3 Tx Simultaneous operation is supported only in AWR1243P with 1V LDO bypass and PA LDO disable mode. In this mode 1V supply needs to be fed on the VOUT PA pin" [note AWR1243P is an automotive (77G) part and SDK does not apply to this part]. So for your evaluation, assume 2 simultaneous tx only at this point.

Looking at Justin and Piyush_'s comments, I am left with several questions. Justin's response implies that BPM-MIMO is supported and available on the 6843. Piyush_'s response seems to say it is not.

(I am working with DCA1000EVM / IWR6843ISK)

My questions:

1.)Does the 6843 chip hardware have the capability to do BPM-MIMO? i.e. is the chip hardware capable of doing simultaneous transmission of BPM signal on 2 or 3 of the Tx antennas?

2.)When Piyush_ says it is not supported what does this mean exactly? Does it mean that BPM is not implemented in any of the labs/demos? OR does it mean that the latest SDK does not allow it? OR does it mean that the 6843 hardware is simply not capable of it?

3.)Does the current SDK, (MMWAVE-SDK 03_01_01_02), support BPM-MIMO on the 6843? Do any previous versions of the SDK support it? Will any future versions of the SDK support it? When might these be available?

4.)If the chip hardware supports BPM-MIMO but the SDK does not, is there any way that a programmer can still use the chip's capabilities and implement BPM-MIMO in an application?

5.)I am interested in BPM-MIMO because I want as much range as possible while still getting the benefit of 12 virtual antennas for angle resolution. Any suggestions on achieving this on 6843 without using BPM-MIMO area also very welcome.

Thanks in advance for any clarification on this

A1: Yes, 2 tx

A2: BPM is not implemented in the out of box demo in the current 3.x versions of SDK, it is being attempted in the next (3.2) release. At a high level there are two things involved in BPM : 1) Configuring the front-end (RF) to do BPM and 2) Signal processing to be able to decode BPM. The reference code Justin pointed out in 3.1 is code that is issuing the MMWave API for BPM configuration, but the demo will not work with this configuration as there is no processing [it may even crash but it will certainly not produce meaningful final result (point-cloud)]. Currently the configuration + signal processing is only implemented in 2.x SDK oob demo on the 1642 device, i.e on 2.x you can actually exercise the BPM-MIMO feature in totality but only on the 1642 device (which I know is not your device of interest). You can use the out of box demo implementation of 2.x as reference if you want to port the changes to 3.x but this may be difficult as 3.x is a new software architecture. If you want to do this, read the doxygen documentation of the out of box demo in the 2.x release to understand what is involved in the data path processing chain for BPM besides configuration of the sensor.

If your goal is only to evaluate the RF transmission character of BPM, then you may be better off using mmwave studio which does not involve SDK/demo.

A3: I believe answered in A2.

A4: Yes, refer to implementation details in SDK 2.x and carry them to the SDK 3.x demo (this may not be easy), or wait until we make it available in new SDK 3.x (most likely the next one which is 3.2).

A5: Don't have other suggestions.

thanks

Hi Piyush_, 

Thanks much this is helpful.  I will have a look at mmwave studio.

Follow up questions:

1.)Do you know when SDK 3.x with BPM supported for the 6843 will be available?  Rough estimate is fine.  

2.)Your A1 says Yes, 2Tx.  This leads me to a basic question on BPM.  Is it even theoretically possible to transmit BPM on 3 Tx antennas and decode in such a way that we get the 12 virtual antenna benefit for angle resolution?   The TI document MIMO Radar reference 3 describes using an example of using BPM with 4 Tx antennas and decoding involves  a 4 × 4 Hadamard code.   I saw this and made an assumption that BPM theoretically could be transmitted on 3 antennas and decoded correctly somehow with 4 Rx antennas.  Is this assumption wrong?  Maybe the nature of BPM requires 2 or 4 Tx antennas only?

3.)On this thread, Vincent references another post:   https://e2e.ti.com/support/sensors/f/1023/t/678806?tisearch=e2e-sitesearch&keymatch=BPM, made by Enric Gardein.  Enric describes a TDM scheme using 3 Tx antennas:  all 3 Tx antennas broadcasting different chirp patterns, but at any time 2 of the 3 antennas are sending exactly the same thing.  This will give 2x the transmit power and still give the benefit of 12 virtual antennas for angle resolution.  It seems to me this scheme could be used on the 6843.  Is there any reason this scheme could not be used on the 6843?

4.)If I was willing to give up angle resolution could I just send the exact same chirp pattern on all 3 Tx antennas on the 6843?   Thus getting 3x the transmit power but no virtual antennas, (only the 4 physical Rx antennas).  Is there any reason this could not be done on the 6843?

Thanks much,

Rob

A1: I am sorry but we cannot provide a timeline for unreleased features on this forum. The SDK release notes and user guide will include information about the functionality when it is released

A2: I think 3 should be possible if a device were allowing 3 simultaneous tx (1243P does, 6843 does not :  ). As long as the tx contributions are separable in the signal processing it should work I think.

A3: I don't see why it wouldn't work on 6843.

A4: Yes for a device which allows simultaneous 3 tx (1243P) but not for 6843 (only 2 tx simultaneously). For 6843, you could pick any two antennas (but same two for all chirps) and no phase change, say both in same phase and you will have the benefit of higher signal power but angle res limited by 4 rx. Note in this case your theoretical SNR gain over TDM-MIMO will be 20log10(2 = numSimultaneousTx)) = 6 dB (this is basically the tx beamforming gain, although beam direction is always fixed -- at boresight), compared to the 10*log10(2) = 3 dB theoretical gain in the BPM-MIMO case of 1642 (and angle res limited by 8 virtual antennas and you can emulate this case on 6843 if you like). In Enric's scheme I calculate the gain will 10*log10(2^2/3) = 1.2 dB, but you get 12 virtual antennas. So these different schemes have different trade-offs between SNR improvement and angle res. If you could do 3 simultaneous tx (possible on 1243P), then the gain/res numbers would be 20log10(3) = 9.5 dB/ 4 rx for tx beamforming and 10log10(3) = 4.8 dB / 12 rx for BPM-MIMO.

Hi Piyush_,

This is very helpful, thanks again.

1.)Regarding A4:  Could you explain in detail, (or give a reference),  how you compute the theoretical SNR gains?  I would like to understand this thoroughly so I can do the calculations myself.  My understanding now is that TDM-MIMO always has only one antenna Tx for any given time slot.  BPM-MIMO with 2 Tx antennas has both antennas broadcasting simultaneously -----> twice the signal power broadcast as TDM-MIMO -----> SNR benefit of 10*log10(2) = 3dB.  This makes sense to me.   However my understanding is obviously incomplete:

-For the case of 2 Tx antennas simultaneously sending identical signals, with identical phase, why do you calculate this as 20*log10(2) = 6dB?  20x factor assumes an amplitude ratio instead of a power ratio?  (I would have thought this would be SNR benefit of 10*log10(2) = 3dB.

-For Enric's scheme where do you get 10*log10(4/3)?  I would have thought this would be 10*log10(2) = 3dB  since we always have 2 antennas broadcasting identical signal at any one time.

2.)I looked at data sheet for AWR1843.  Looks like this part can do 3 Tx antennas simultaneously.  Is this correct?  

3.)Does AWR1843 support BPM-MIMO?  If not is it planned to have support for 1843 BPM-MIMO in future SDK's?

Thanks much,

Rob

Hi Piyush_,

One more question:

Are you able to quantify the number of added meters of range we will get for an SNR gain of X dB?   e.g. if BPM-MIMO gives me a 3dB SNR gain over TDM-MIMO, then how many meters of range am I gaining using BPM-MIMO?

Also please relate this antenna radiation pattern, (which I know is not uniform), and to the range of angles our antennas are covering.  Can you give a brief explanation of the key ideas and perhaps references for me to look in more detail?

Thank you,

Rob

I meant to say can you please relate this to antenna radiation pattern. i.e. if BPM-MIMO gives 3 dB SNR gain over TDM-MIMO, what is the added range in meters we get over the angular range covered by the antenna radiation pattern?

On the question of SNR calculations:

Reference single tx [or tdm-mimo]: rx = s + w (noise), signal of amplitude A has power A^2, say noise power is Pn [= E[w^2]) so SNR = A^2/Pn = X

Tx Beamforming rx = s1 + s2 + … sN + w, if each s(i) has same amplitude A, they add-up in the beam direction constructively, signal power =(NA)^2, noise power is Pn, so SNR = N^2 * X, so gain over single tx is 10*log10(N^2) = 20*log10(N)

BPM-MIMO: In order to get the virtual antennas we have to separate (detection happens after separation), after separation in receive processing, rx = s + sum(w)/N [here sum(w) may have +/- combinations], signal power = A^2, noise power = (1/N^2)*E[ sum(w)^2] = sum(E(w^2)) [cross terms expectation = 0 because uncorrelated noise] = 1/N^2*(N*Pn) = Pn/N, so SNR = A^2/(Pn/N) = N*X, so gain is 10*log10(N)

Enric's scheme: Same analysis as BPM-MIMO except rx = s + sum(w)[3 terms]/2 [see equations in his forum post]. So SNR = 2^2/3 X.

Here noise is assumed to be uncorrelated, so for max range purposes the noise assumed here is the receiver (thermal) noise. I guess the R^4 relationship means SNR improvement of Y dB will result in max range improvement of Y/4 dB, everything else being equal.

As far as antenna patterns are concerned, in BPM-MIMO (or Enric case), even though antenna patterns are different in the different chirps (unlike tx-mimo where they are same if we assume all tx's have the same antenna pattern), after separation we separate each tx contribution so effectively it looks like we were doing TDM-MIMO and therefore I think it will not really change anything fundamentally compared to TDM-MIMO. In the separated domain, compared to TDM-MIMO, we can see that the signal power is not getting enhanced, rather the noise power is getting suppressed by 10*log10(N). Situation is different for tx beamforming, in this case there is no tx contribution separation, so this case, the tx beam matters and it is more directional compared to single tx (narrower beam but tx power is more). Ideal case of tx beamforming will have spacing of lambda/2, not the 4(=numRx) * lambda/2 = 2*lambda spacing we have to create the virtual array for TDM/BPM-MIMO [this is the point you brought up earlier our EVM pattern is not for tx beamforming but a real product will design antenna arrangement to their requirement].

1843 is not an industrial part, it is automotive part, not sure if you can use it for your application (you mentioned surveillance which is industrial application). Parts that begin with letter "I" are industrial (60 GHz) and parts that begin with letter "A" are automotive (77 GHz).