09 March 2012

splitter math and hands on

Hands on first with some pics, then the long wordy text after the jump.

I did some quick 2.4Ghz band checks with splitter inline, without, and then with the splitter+LNA inline and switched on.

These were done in the same manner as the previous traces for my LNA Functionality blog post.

A quick description is:
Cable an Andrew DL-2402 antenna to the spectrum analyzer input and adjust so the 2.4Ghz band is roughly full screen. Then run the spectrum analyzer for a couple minutes in Max Hold for trace A, then freeze the trace, leaving it onscreen. Then run Max Hold for trace B with another set of connections, eventually freezing that one after a couple minutes.

(See the LNA Functionality post for a more detailed explanation of how the traces are acquired.)

Here's the Splitter vs NoSplitter test setup:

NoSplitter setup shown, already did Splitter trace

Closeup of the screen from picture directly above. I must have had the video bandwidth turned up more here than in the next test setup so these traces look less fuzzy/more smooth. But it's still valid to compare the two within this picture to each other:

2.4Ghz band Lower trace is splitter inline, upper trace, no splitter (10dB/div)

This is the full setup, antenna into LNA, into splitter, into spectrum analyzer:
Test setup for splitter+LNA vs splitter only(shown in splitter only setup)

Closeup of screen from picture directly above.

2.4Ghz band. Upper trace is splitter+LNA inline, lower trace is splitter only (10dB/div)

Basically that last picture there is the "proof" that the LNA into the splitter can work pretty much as well as a single antenna into a single radio for an RX only setup. I don't have a comparison of splitter+LNA to Direct-to-Antenna. I don't' know how i forgot that but... you can look at the LNA Functionality post and see a direct to antenna trace for comparison. (Don't compare absolute position on screen, I'm not sure things were all referenced the same between those tests and these ones here.)

I'll put the wordy rambling about splitters after the jump where I'll talk about
  • Perfect or Ideal Splitter Math
  • Splitter Reality

Perfect or Ideal Splitter Math

The splitter (as I refer to it) is also called a power divider. Think of it like dividing a gallon of milk, you now have 2 half gallons. Split each 1/2 gallon and you now have four 1/4 gallons. There are no magic calculations to be made. If you split it(evenly), you have half on each side.

There is such a thing as an uneven split (in certain situations this may be preferred), but just remember: If you split 1 gallon of milk the two(or more) portions created have to add back to 1 gallon. You can't split 1 gallon into three 1/2 gallon and one 1/4 gallon portions... that adds up to 1.75 gallons.

In RF decibel measurements, 3db is either a doubling or halving of the power depending on whether you are adding 3db or subtracting. If you split an RF signal (evenly) you get two "pieces" each 3dB lower.

Another analogy:
Take a 5kW generator, you can put on a power strip to split the power it generates to multiple appliances, but you are still only going to get 5kW out of it. You can use the 5kW of power in whatever "pieces" you want, a 1.5kW hair dryer, 20x 5 watt cell phone chargers at some tiny fraction of a kW or whatever, but you can't get more than 5kW.

Anyway, to get to the point: a 16 port even split power divider is going to have 12dB of loss. There is no way around that anymore than you can divide 5kW of  generator power into 8 individual 1kW loads to achieve 8kW of power.

Let put it in some actual dBm numbers(click image to expand):
16 Way even split - 4 Tiers of splitting, -3dB per "tier"

Splitter Reality

You can refer to this datasheet (for the splitter I'm using) ---> Mini-Circuits ZC16PD-2185 Datasheet

In reality the splitter follows theory pretty well, but there are some internal losses so you aren't talking about exactly 12dB, but ~12-13dB can be expected. The manufacturer should spec the losses for the splitter. They may spec them referenced to "theoretical", so for a 16-way 12dB theoretical splitter, they may say the loss is <1.1dB per port or something. Meaning 12-13.1dB of real loss.

The datasheet linked above clearly describes the losses as above 12dB, some may not.

ZC16PD-2185 datasheet clearly describes losses(also includes a example measured on all ports not shown here)

Everything above is referring to the RX direction, into the sum port of the splitter and out the branches. I'm building an RX only system, but unless I can find a way to solder/hack in a TX disable it's not impossible or incomprehesible for the radios to TX into the splitter branches. Some wifi chipsets may do some automagic scan/probe when it powers up. Someone may know better, but I have no idea.

Generally, they splitters will work as combiner to the main (or sum) port, which is connected to my antenna. There's no harm in that. What I am concerned about is how isolated the 16 branch ports are from each other. I don't want an accidental TX on branch 8 to come back out branches 1-7 and 9-16 with enough power to burn up their RX.

The splitter datasheet specifies the isolation between branch ports as 30db typical, 16dB min. If you  assume 20dBm max out my usb radios(I haven't' looked into what it really might be yet) you could say that i might see +4dBm worst case at another radio input, and -10dBm typical.

Isolation spec from datasheet for 16 port splitter
There's no doubt things won't work in that state, the RX would be blown up or swamped. I do have confidence that I can mute the transmitters once they are powered up and monitor interface created. So my problem becomes only whether the radios can survive inadvertent or accidental TX by the others.

I'll be testing it out soon with just a couple radios. I have a few things going for me:
  • I'm going to configure the radios only interface in monitor only (iw phy phyN interface add wlanN type monitor)
  • I'm only running kismet on openwrt, no NetworkManager or wicd or injection stuff
  • I've made and tested an rt2800 driver patch that skips automatic wlanN device creation (so I can create a single monitor type interface only, further VAPs then can't be sta/ap etc I think)
  • I don't think the cheap radios I have actually put out +27dBm as reported by iwconfig. Amp-meter measurements show only ~70-90mA of current at 5V, which is .45watt, so unless it's converting 100% of input power to RF power... it's not going to put out +27dBm(+27dBm IS .5W) I did some limited amp testing in a few machines, OS'es and with/without external power but the best I could t it to use was 100-110mA intermittently during TX.

Worst case, I have to figure some way to disable or dial down the TX's or install an isolator (one-way RF device that has 20 or 30 dB of loss in the opposite direction) between each radio and splitter branch to pad down the inadvertent TX'ing enough.

I did power up one of the radios and was not able to detect any RF out with the spectrum analyzer, but.. it might have been such a quick blip that the relatively slow sweep just missed it. I tried a few adjustments and several radio power up attempts but I was not able to see any RF. That was on openwrt with my patched rt2800 driver that prevented interface creation.

I did not try it while powering them up on Fedora with NetworkManager etc.

I'll revisit the no-TX issue after I get things all mounted and built up hardware wise. I need to look into RFKill stuff and if that can provide any no-TX assurance.

No comments:

Post a Comment