Upcoming Maintenance Alert:

The UBNT Community will be upgraded at 5pm MDT on April 25th. During this time the community forums will be set to read-only status.

Learn more

×
04/17/2015
Radio Shootout Pt 1 - AC Lite vs. Brand M
Description

Mim-AC.JPG

I was initially going to compare the AF5X with the Brand M radio (in the picture above) so that everyone could see how the two really compare.   But once I did some baseline throughput testing on the bench, it became obvious that Brand M wasn't really in the same league as the AF5X, and was really more like the AC radios (which is actually what it is - it even says "a/n/ac" in one of it's wireless mode selections).   Below is a quick look at the max speeds of a RocketM, and AC Lite, an AF5X (actually on early firmware, so the ones people are getting now are actually faster) and Brand M.   This is a quick and dirty baseline test - the tests farther below were much more thorough and better set up, but all the radios were treated equally for this first one.

BaselineShootout.png

What became immediately obvious was that except with an 80MHz channel the Brand M radio, comparing apples-to-apples (with just one of it's 2x2 radio channels operating) is significantly slower than the AF5X, and still slower than the RocketAC Lite.   

So at this point I abandoned the idea of head-to-head testing (at least for throughput) with the AF5X, which really outclasses the Brand M, and instead stick with the ACs.

 

One thing that should be finally put to bed here - if anyone out there still thinks there's some magic going on inside the Brand M radio for it's "160MHz channel" other  than effectively two AC radios in one box, here's the proof - an AirView of the radios actually running:  

B5cDualRadio.png

If you look at the areas directly adjacent to the two 40MHz channels, you can see how much IM junk is generated in the PA - this is on the bench, so there's no interference from outside radios.

 

Anyway, after this I just concentrated on comparing it to an AC Lite.   Following is a series of automated tests comparing the AC Lite and the Brand M radios.   These are all using iperf in tcp mode, with 10 streams each, and with a complete tests running in each direction at once, so this is testing actual throughput both directions simultaneously for each radio.   Note that the PHY rate (dotted line) is the actual theoretical hardware encoding rate possible at that MCS and channel width, and the actual throughput of the radios is always lower than that - often much lower.

Throughput_bidir_R5AC-LITE,MIMO-B5C,['1x20']1429223748.pngThroughput_bidir_R5AC-LITE,MIMO-B5C,['1x40']1429223942.png

Throughput_bidir_R5AC-LITE,MIMO-B5C,['1x80']1429223960.png

You can see that in 2x2 mode at 80MHz the Brand M is faster than the AC Lite - at that channel width, the AC Lite is at the edge of it's chipset's capabilities, plus it was designed to do more narrow channels.   And who has clear 80MHz chunks of spectrum anyway?

 

Since the Brand M radio has the capability (as seen in the AirView above) to run as two 2x2 MIMO radios on two different channels, and combine the data streams together, the analogous setup for an AC radio is to use two discrete radios, each on it's own channel.   So this was done also (the setup actually used an EdgeRouter in load balancing mode to create the two streams for the two radios, and then another one to recombine them on the other end):

Throughput_bidir_R5AC-LITE,MIMO-B5C,['2x20']1429223709.png

You will note in the above test that the Brand M radio shifted speed part way through the test.   I saw this kind of behavior in other areas too - changing power output to see what effect that had on throughput, I found I could never quite get back to the performance level I initially saw.   And some times the performance would drop off after the radio heated up - this may be indicative of some thermal management issues in their power amplifiers - the actual quality of the PA , or maybe they just need a bigger heatsink...

Throughput_bidir_R5AC-LITE,MIMO-B5C,['2x40']1429223662.pngThroughput_bidir_R5AC-LITE,MIMO-B5C,['2x80']1429223987.png

Note that in the dual 80MHz case the AC Lites are faster than Brand M, unlike in the single 2x2 test before.  My guess is the PA is reaching the limits of it's IM cababilities trying to linearly amplify the two separate frequencies, and therefore the modulation rates go down.

 

Since the test was already set up, here are the AC Lite's results for all the other channel widths it can do:

Throughput_bidir_R5AC-LITE_1429224656.png

And here it is with dual radios:

Throughput_bidir_R5AC-LITE_1429224655.png

Now it's legitimate to ask if using two complete AC radios and an EdgeRouter Lite on each end is a fair comparison in the real world.   Well, economically, a pair of Brand M radios costs $1678, vs $740 for 4 AC Lites plus two EdgeRouters, a less than half the price of the Brand M radios alone.   And yes, to make this work you'd need 2 more antennas for the ACs, but this is a pretty corner case solution anyway - where but in the middle of nowhere or over a very short link are you going to be able to do this (especially at 80MHz channels) and in those cases you can probably do better with AF5s or AF24s anyway.

 

So what's the upshot of all this?   Well, things don't look so good for the Brand M radios so far in this shootout...

 

Part two will look at some other radios (plus these two again) and how they fare in interference situations. 

Jim

Comments
by
on ‎04-18-2015 09:40 AM

Great Write Up! But I need to see results at 34 miles.

by
on ‎04-18-2015 09:56 AM

Nice,

 

Regarding AirfiberX,  is seams to me that this product is more equivalent to Cambium ptp650 !?

I dont have experience with neither of them, but that test woud be nice

by
on ‎04-18-2015 11:13 AM

pcmcnary - don't have a link that long, but we are going to be testing their radios against AF5Xs on a 21 mile link later this week.

 

Makoto - Yeah, the ones based on the  old Orthogon radios - don't have one of those to test either...

Jim

 

by
on ‎04-18-2015 11:40 AM

What was the framing of M 2/4/8 ms?

You can't compare the frequency reuse case in lab.

M has working GPS Sync. So you can use bigger channels

as you might use them 3 times at one tower.

by
on ‎04-18-2015 12:11 PM
 
by 28m ago

What was the framing of M 2/4/8 ms?

You can't compare the frequency reuse case in lab.

M has working GPS Sync. So you can use bigger channels

as you might use them 3 times at one tower.

 

The test was for 4ms    And the AF5X has GPS sync, which can make channel reuse of as many radios as you want.   And we're doing that in the field right now.

Jim

by
‎04-19-2015 03:34 AM - edited ‎04-19-2015 08:27 AM

These tests show max throughput vs channel bandwidth and MIMO streams  at 256QAM modulation, that is measured  in lab.     Average result – it is about  70MBps MIMO 2x2 simplex in 10 MHz channel bandwidth for 802.11АС . Is is not new information. 

 We   see  new, that AF5X provides   max about 85 MBps simplex in 10 MHz channel.  Mimoza  AC radio AC ( it seems it  is latest  wave 2  AC chipset MU-MIMO) provides max 60-70Mbps MIMO 2 x2 ( 2 streams),  that is less that rocket AC.

But it is max throughput. What will  we get in real word?

2 jim , you have set up with interference sourse. Provide tests with co-channel interference.

We already have seen  real links  on  Powerbeam  АC  at 12 km distance LOS( 40 MHz BW, RSSI -57 dBm,  CINR= C/(I+N) =30 dB,  256QAM ). PB AC is not able to give more that 160-180Mbps TCP aggregate ( UL+DL)  at  duplex ( two way) traffic, without latency degradation above avg 20 mc.  So  max 70 Mbps  simplex   in each 10 MHz channel bandwidth ubnt AC in lab condition became 40-45 Mbps (UL+DL) in  each 10 MHz in real links.

I suppose that MU-MIMO chip  ( that  is probably used  in Mimoza)  has better BER at 256QAM and lower CINR  value, that  is required for 256QAM, than AC wave 1 ( rocket AC).  I think  that in spite  of  new AC chip gives slightly less throughput in lab, it may be able to keep high  speed in real links. Please test Mimoza  and Rocket  AC at 256QAM  at  CINR 30 dB ( when required CINR level for 256QAM  is 34-36 dB).

Also the similar test  will be interesting with AF5X. I suppose, that AF5 radio and protocol is not able to resist enough against   interference and multipath ib 5 GHz. I suppose that AF5X throughput in real word may l be in twice or more worse than in lab. You may check it just now,  please, do it.

by
on ‎04-20-2015 07:10 AM

by
on ‎04-20-2015 09:38 AM

Am I misreading the data which shows the RocketAC being faster than the AF5X in terms of throughput vs channel size?

by
on ‎04-20-2015 10:52 AM

@wispThe AirFiber and Brand-M radios use a version of TDD where the airtime for upload and download are fixed in size. AirMAX uses TDMA which allows for dynamic airtime allocation.  This means that in one way speed tests the AirMAX radios can dedicate virtually all of the airtime to that one direction, but the other radios are restricted by their settings.  This is why eejim added the aggregate column on the end of the tables for the AirFiber and the Brand M radios.

by
on ‎04-20-2015 11:11 AM

Thanks for that info.