You are here

split MIMO

26 posts / 0 new
Last post
KE2N
KE2N's picture
split MIMO

As an interim measure, where we want two directional antennas, pointing in different directions on 2.4 (on the same channel) that we could connect one "chain" of the radio (M2) to one of the inputs on a MIMO antenna and the other chain of the radio to a single-polarity antenna pointing in the other direction.

The idea is that, later, the node would be converted to two-frequency operation using MIMO in both directions.

Has any one done this? It seems to me that the radio does not know what it is connected to and will just operated in MCS 1-7 as if one of the polarities of a standard MIMO setup was blocked.

 

K5DLQ
K5DLQ's picture
I think KG9DW failed
I think KG9DW failed miserably when he tried this.
OMHO, the issues I think that you would run into are: no shielding/isolation (since it is single frequency) and the hidden transmitter issue become MUCH more likely.

I'll defer to other for additional comments.

 
kg9dw
kg9dw's picture
Failed miserably is an
Failed miserably is an understatement. It was a big waste of time and money. Ubiquiti and other mainstream suppliers build antennas optimized for WISPs, and I've had great success sticking to those antennas. 
AE6XE
AE6XE's picture
In theory:

[correction for future readers]

In theory:

1) the MIMO device has a transmitter on each antenna chain and will send the energy in different directions.  The transmit throughput would/should always be much worse if splitting the power and trying to send half the energy in the wrong direction (and would never be able to do 2 spacial streams with 802.11n MCS8+ rates).
2) On receive, the MIMO device and this design could run into trouble with the Atheros Ambient Noise Immunity (ANI) algorithm in the software driver.    I have seen the ANI subsystem manipulate the SDR such that one of the antenna chains essentially becomes deaf as the algorithm tunes in a signal on one antenna and attenuates the other antenna to shut out the rift-raft.   This is very undesirable when there are single antenna neighbors now dropping out on the polarization that goes deaf.  We put in some checks to get out of this state in 3.15.1.0.  

Consequently, you might give it a try and it might work 'sufficiently', but then it might fail miserably as noted (but, before 3.15.0.1?).  This is fairly complex in the drivers how all this works.  It's difficult to know what to expect until we try it and get some data--for a use case outside the intended design.   

Joe   

KE2N
KE2N's picture
H and V

What does the MIMO radio do if it has one "customer" that is horizontal-only and another "customer" that is vertical-only?  
I would really expect to handle that ... and the radio would see essentially the same thing in the setup I proposed. 

KG6JEI
Yes and no, there are
Yes and no, there are multiple facts that come into play here.

With differing polarity there is a 30db isolation between signals while in your method is the two antennas share the same polarity there would not be a 30 DB difference.  This comes into play as signals bounce and you end up with multipath to the indirect antenna which could degrade the decode and since it's not attenuated by a second polarity it has 30db more to work with (the chips assume they are seeing same physical space so the math between the two signals assumes the RF is gonna take the same path.) this is as compared to the verticals and horizontal user being isolated from each other. In both cases the unit will lock on to one but it may not be the one you want (causing a bad packet error and an eventual retransmission)

As Darryl Mentions you now also have two directions of RF open to you to jam the receiver.

Ultimatelu yes the hardware will function but it wasn't designed to work that way and performance can be expected to be degraded.

 
KE2N
KE2N's picture
orthogonal

I guess I neglected to mention that the two antennas will be directional ones and pointing at right angles to each other - so I would expect a fair bit of discrimination between the two signals. But you make a good point.  One of the signals is vertical. At this stage, we could arrange for the signal(s) from the other direction to be horizontal, during the interim phase, if that would be what is needed for success.

AE6XE
AE6XE's picture
Ken,  I suspect you are using
Ken,  I suspect you are using 2 x dual polarity antennas.  1 antenna is on the transmitter chain A on H-pol, the 2nd antenna is configured V-pol on chain B of the transmitter, correct?  
KE2N
KE2N's picture
antenna details

Short answer: yes, I think you have it right.

Long answer: The antennas are not bought yet. We are thinking to use a dual-polarity sector antenna with V (or H) connected to one chain (and the other port of the antenna not connected, or terminated in a load).

In the other (perpendicular) direction - and connected to the other chain - the antenna would be vertical-only and a uni-directional/linear type such as a Yagi, or a small reflector.  

The transition plan is to go conventional MIMO on the 2.4 sector and replace the uni-directional link with 5 GHz equipment (also MIMO).

That uni-directional./linear-pol antenna would be no longer needed after the transition - so it will be whatever we could get cheaply (or free). 

The other (i.e., conventional) way of doing this would be to use a single low-gain omni MIMO with no down-tilt. That would catch both of the signals.  But it would be at a 10 dB disadvantage RF-wise for the mesh coverage (say 6 dB versus 16 dB) and we would not need that antenna after the transition, so it would be a waste of resources.
 

KG6JEI
Incorrect in point one. The
Incorrect in point one. The unit DOES NOT try and focus the energy to the single chain.  BOTH CHAINS receive equal energy. The system does not support BEAM FORMING which is the process of directing power. The only difference between lower and higher rates is the splitting of data between the two chains but the same amount of RF power is used.
AE6XE
AE6XE's picture
Conrad,  not clear on what is
Conrad,  not clear on what is incorrect on point 1--which is only about transmit?    In the proposed configuration, only modes MCS0 - MCS7 would be selected and the correct antenna chain would be transmitting to the neighbor in that direction (barring really good reflections off the other antenna pointing in another direction).     The MCS8-MCS15 rates (which split the power across 2 antennas) would never be selected to transmit to a neighbor as half the energy would be going in the wrong direction and yield less throughput--we can confirm this by looking at the rate selection rate table for the given neighbor. 
KG6JEI
MCS0-7 transmits the SAME
MCS0-7 transmits the SAME DATA out both antenna ports. Both chains transmit at once. MCS0-7 does NOT shut down one of the ports. The device does NOT "try and focus the power" (that's called beam forming in wifi land).

Hence the same amount of RF power always exists per chain no matter if it's a single data stream or dual data stream (ignoring that the radio decreases power for faster speeds the pairs rates are same power aka MCS0 is same power as MCS8 and MCS1 same as MCS9 ).  So "half the energy" is ALWAYS going off in the wrong direction.
AE6XE
AE6XE's picture
Conrad,  Yea, I thought in
Conrad,  Yea, I thought in past readings that these Atheroes chipsets are able to do transmit antenna selection diversity (beam forming is much more, playing with phase and power proportions across multiple antennas).     The link layer in the driver has all the information to tell the chipset to select a transmit antenna for a given neighbor.  But yea, if the chipset can't do it or the logic is not in the driver, then no path from A to B.    

So if this is correct and the transmit energy is always split to the 2 antennas, back to the scenario with the antennas pointing in opposite directions, the SNR at the receive end is down 3dB and missing one polarity (and roughly equivalent if it was a single antenna device neighbor--which isn't receiving much energy from the off polarity signal anyway). 

This also means and would quantify and put into perspective that any mixing of a bullet with a nanostation puts the bullet down ~3dB SNR receiving compared to if 2 Nanostations are used.   

Joe AE6XE
n1cz
n1cz's picture
I am not totally up on mesh
I am not totally up on mesh net, but what I think I understand is a node may as well be considered isotropic? ... if mimo? And, mimo is a requirement if only 1 channel is available. But, the "other" site may be anywhere that sufficient signal is present to both locations. And I could type more, but most understand the mesh ability to reconfigure.

I once coordinated point to point microwave repeater sites so I think in term so of cross-polarized antennas with co-located transmitters on adjacent channels. That's the simple case. No reconfiguring. If the link breaks it is broken unless there is some kind of diverse routing designed in via hardware.

"... am I right?!" At least to a point
AE6XE
AE6XE's picture
I would use the word "multi

[slight update to account for 1 spatial stream splitting energy out 2 antennas]

I would use the word "multi-point" and customized transmit to each neighbor to better describe the mesh node characteristics.   The antenna radiation pattern defines the coverage area of the multiple links that may establish if other nodes are in view.   We're using 802.11n (well, could also link to a node if it only did 802.11a/b/g) in "adhoc" mode, which means CSMA:   https://en.wikipedia.org/wiki/Carrier_sense_multiple_access_with_collision_avoidance

Transmitting:

These SDR radios, the Atheros chipset and the software driver in use, will discover the best thoughput option to transmit data to each neighbor uniquely.  It does this by trying the RF options available to transmit to each neighbor.   Packets going to different neighbors go at the best throughput discovered/measured for that neighbor.   Every other packet can be transmitted using a different modulation type (64-QAM, 16-QAM, QPSK, BPSK), forward error correction code, and spatial streams of data  (for a MIMO transmitter with 2 antennas using 802.11n, the combinations are spelled out as MCS0 to MCS15).

In this scenario of pointing an antenna in one direction and another antenna in the opposite direction, the radio will discover that the best throughput is using only 1 spatial stream and send half the power in the wrong direction and be roughly equivalent if the receiving neighbor was a single antenna device not getting any energy from the off polarity.

Receiving:
 
The receiver will combine the energy from the 2 antennas using Maximum Ratio Combining (MRC) -- wiki articles out there, but basically the energy received by both antennas is combined to obtain the best signal to decode (given 1 spatial stream).   With OFDM, this is done in the freq domain so that phase issues can be ignored and don't cancel out the signal (if they aren't already canceled  out in the environment before getting to the antennas).    The chipset provides 'I' and 'Q' values.
   
There's some ability with this ANI subsystem to adjust various parameters independently on each chain--it can be observed that one antenna chain can be attenuated by itself if the algorithm thinks this noise is causing problems (minimizes error rates).    ​

MRC should not be confused with antenna 'selection diversity' or 'switching diversity' which basically says, pick one antenna to receive on.   I've not dug in sufficiently to uncover if Atheros and the software driver will choose between diversity and MRC, or just does MRC all the time as an improved technique for all conditions.  What it looks like is that MRC is used all the time, but the ANI subsystem may adjust the parameters to essentially attenuate out one of the chains to essentially be doing selection diversity  (and I saw this occur for all signals coming in and not unique to a given neighbor).

Back to pointing the 2 antennas in opposite directions.   The mesh node still has the issue of CSMA from all the signals--this doesn't change. The difference here is that one of the two antennas becomes ineffective at receiving the signal from a given neighbor.  That neighbor would not be transmitting 2 spatial streams if it was a MIMO device and would stay at MCS0 to MCS7 (unless the devices were really close or had a good reflection path).     The SNR will simply be less when one antenna is not aligned receiving from the neighbor.   I think a given environment would just have to try it to determine if the SNR is still sufficient (with any additional noise coming in on the unused antenna chain). 

Joe AE6XE

KE2N
KE2N's picture
it sounds promising

I am comparing this with a single low-gain OMNI which "radiates equally poorly in all directions" as the saying goes.  Losing 3 dB due to the split, while gaining up to 10 dB for the higher gain antennas is a net win.

I would point out that transmitting on the chain that is pointing in the "wrong direction" is in fact useful.  It reduces the hidden transmitter issue by telling the other station to hold off and not transmit at a time when the mid-point station would not be listening anyway. (Contest stations with multiple antennas use this technique all the time: they listen on one antenna but, transmit on both/all to keep the frequency clear). 

Anyway - without a huge effort I think I can stage something on the ground here with two nodes that cannot hear each other and a common node in split MIMO mode.  It might be a couple days before I report back.  
Stay tuned

kg9dw
kg9dw's picture
I welcome your experiment.
I welcome your experiment. Having tried to do it in production and having failed to meet any of my objectives, I'm very skeptical. If you need multiple highly directive point to point links from a single site, the proven way to do this is multiple point to point MIMO radios on different frequencies or even bands linked together using Dtd. Why cobble something together for $50 when you can add a complete 5GHz parabolic node for $100? 
KE2N
KE2N's picture
that's the plan

I agree.
That is the long range plan.
But there is no 5 GHz at the moment. So we need 2.4 going in different directions and elevations. 

KE2N
KE2N's picture
testing complete and it works.

I finished the tests yesterday. The setup is shown in the attached PNG files. I included quite a strong "competing" signal for a touch or realism. The wanted signals  were moderately weak signals (MCS 2 or 3 mostly) and, using the rule-of-thumb of 20-40% of the bit rate for throughput (halved because two hops), I figured 2.2 to 4.5 MB/s should be the expected result.  

My results (using “iperf”):

  • ​Two radios – 3.8  +/- 0.4 Mbps
  • One radio split – 4.2 +/- 0.7 Mbps

Since the orientation of the sector antennas probably changed slightly between the two tests (introducing an uncontrolled variable) I would say that these two results were essentially the same.
I ran the setup for several hours and several hundred MB of data transfers (in both directions) without any problem.  (The only problem I had was when OLSR decided my airRouter should connect to one of the remote units directly - instead of to the much stronger mid-point node - but that is another topic altogether. I should have used 5 GHz for the PC link.)

Image Attachments: 
VE3BYS
VE3BYS's picture
2 Radios with the DTD Link
The above is a very interesting situation.

Do you think I can do the same thing but in a slightly different way by getting rid of the DTD link and having 2 radios? That is one radio each connected to the V on both antennas and the other radio connected to the H on the other antenna? This will get rid of the DTD link and still make the link between the two dishes. 
 
K5DLQ
K5DLQ's picture
Interesting data Ken.
Interesting data Ken.
KA5TYW
Rocket

I have a Rocket M2 and an AMO-2G10 MIMO antenna. I want to use my AMO-2G10 antenna for vertical polarization omni, only, operation while using a parabolic antenna for horizontal polarization. If this is possible, which connector of the AMO-2G10 should be connected to which connector of the Rocket for vertical polarization, only, operation?  Or would it be better to use a conventional single vertical omni antenna for vertical polarization, rather than the AMO-2G10 antenna?

Thank you & 73,

Paul -- KA5TYW

K5DLQ
K5DLQ's picture
This configuration is not
This configuration is not recommended.  The EXTREMELY VALUABLE part of the MIMO technology is to have it use both antennas/polarities to bring in a better signal and/or reduce the noise.  You really need two radios to do what you are asking (in an optimal fashion).

 
KE2N
KE2N's picture
try it and see (edit)

I agree it is not optimal. But if the other end is not MIIMO what is the benefit of your end being MIMO?

Use of MIMO radios / antennas with single polarity "customers" can be a liability.  We had a case recently using an M2/MIMO sector to link to a single polarization node.  It did not work right until the horizontal channel was disconnected (and capped with a resistor).  The problem likely caused  by a channel 6 signal as strong as -40 dBm in the horizontal channel.  Who says your neighbors WiFi is only vertical.  (And where did he get that amplifier?)

The problem with co-located radios on the same frequency is being discussed in another posting.  The split MIMO arrangement neatly side-steps that particular problem [edit] as the two "channels" ALWAYS transmit at the same time and they ALWAYS receive at the same time - so that you do not have self-interference as you might have from two radios back-to-back.

KG6JEI
"I agree it is not optimal.
"I agree it is not optimal. But if the other end is not MIIMO what is the benefit of your end being MIMO"

Signals can bounce around as they travel.  What leaves you vertically may arrive horizontally at the destination (and vice versa) having both polarities on one side can nullify some of this. Diversity receive gives the node a second chance to pull in this information.

We have seen it time and time again that replacing an AirGrid (which has more gain than a NanoStation) with something like a NanoStation can actually improve links because of the dual receivers on one side.

Two NanoStations have even performed better than two Air Grids (and better than single NanoStation and AirGrid) even though there is less link margin.
 
KE2N
KE2N's picture
V&H

If your plans are for eventual MIMO operation, then pre-mounting the MIMO antenna makes sense.
You might want to terminate the unused port in a 50 ohm load for now.

I checked the AMO that I  have and the vertical is on the left from the viewpoint of the radio (see photo) - this is the same convention used by the L-Com HG2410DPU which has the letters V & H embossed on the base.  I would note that the L-Com has no down-tilt, whereas the Ubiquiti does have electrical down-tilt.  Depending on how high your antenna is, one may be better that the other.  

One thing you need to be aware of is that the OMNI will pick up signals from all directions and this may impact on the sensitivity of the dish antenna.  For this arrangement to work well, you should do a WIFI scan (20 MHz) and check that the OMNI is not picking up signals stronger than about -50 dBm on other channels (1-11).  De-sensing dues to strong unwanted signals is not unique to this arrangement, but use of the omni (or even a broad sector antenna) does give you more chance of running into this.
 

Image Attachments: 

Theme by Danetsoft and Danang Probo Sayekti inspired by Maksimer