Jack,
Even 1, 6 and 11 won't work. The selectivity of the radios isn't sufficient to allow adjacent channel operation without a significant reduction in apparent sensitivity. This will impact range, or negotiated rate, or both.
In fact, making 1 & 11 work without impacting range is pretty difficult. Nearly impossible without a lot of additional filtering, in fact, and, basically impossible with common 11g cards (but see below).
I suppose the maximum usefull number of miniPCI sltos I could see would be six:
card 1: centered at 5.200GHz (uses the 5.150 - 5.250 'UNII-1' band and associated limits, including indoor only operation)
card 2: centered at 5.300GHz (uses the 5.250 - 5.350 'UNII-1' band and associated limits)
card 3: centered at 5.520GHz (uses the 5.470 - 5.725 band, requires TPC and DFS, EIRP limits are like 5.250-5.350, note the restricted band between 5.350 and 5.470)
card 4: centered at 5.620GHz (same issues as card 3)
card 5: centered at 5.800GHz
card 6: 2.4GHz operation
I've not burped (badly) about the 4 port adapter, because someone could use 3 11a cards (or 11g/a cards in '11a' mode), plus one card operating in 2.4GHz.
I picked the center of the band on cards 1, 2, and 5 (6 as well, but not shown) such that a little more antenna gain can be used (.vs at the band edge). Note that the restricted bands (below 5.150, above 5.850, and between 5.350 and 5.470) have strict limits, so additional antenna gain (which, without a really good channel or bandpass filter *will* raise the sidelobes), and therefore will make certification (or legal operation) very difficult.
Thus, card 3 is run 50MHz up from the band edge and card 4 is run 50Mhz down from the band edge.
minimum card-to-card center frequency spacing shown is 100MHz. Minimum (IEEE) alternate channel rejection specs (thats 50Mhz away) is 32dB, (and this is the 6Mbps figure. Also note that the minimal spectrum mask for OFDM modes in 802.11 is some 60Mhz wide (so 30Mhz away from the center frequency) at -40dB. You're not going to run 2 802.11 radios in the same band and not have any transmit that happens by one (including beacons) not destroy any concurrent uplink packet by the other.
Consider also that any directivity between co-located antennas makes the problem worse. Side and backlobes count too.
Given the above, and considering out of band emission limits, and the lack of selectivity in the available radios, no more 'channels' can be used. You basically get one radio per band.
Note that DFS+TPC is coming to 5.250-5.350 as well. The following values are FCC-required for DFS to allow successful "spectrum sharing" at 5 GHz:
1. A DFS detection threshold of -64 dBm for WAS devices operating at a total effective isotropic radiated power of 200 milliwatts to 1 watt, and -62 dBm for devices operating at an e.i.r.p. of less than 200 milliwatts, measured over a period not to exceed 1 microsecond, as normalized to a 0 dBi gain antenna. These measurements must be accomplished during quiet periods between or within each WAS frame or packet;
2. A channel non-occupancy period of 30 minutes to ensure that fixed radars will be protected for any channel in which the DFS detection threshold has been exceeded;
3. A channel availability check time of 60 seconds upon initial startup or monitoring of WAS on any particular channel to ensure all radars present around a WAS are detected prior to it utilizing a channel; and
4. A DFS Channel Move Time of no more than 10 seconds. DFS Channel Move Time is the period that WAS systems will have to move off of the channel once the DFS detection threshold has been exceeded. It takes an average of 200 milliseconds for all normal traffic to suspend and then intermittent control signals can continue for up to 10 seconds.
#1 means that if you find another signal that is -64 dBm or higher, you can't operate on that channel, and you must move.
#2 means that you can't re-attempt to operate on a channel for 30 minutes after you've detected any signal above -64dBm. You have to sit and listen if you choose to retry.
#3 means that you have to sit and wait for 60 seconds (listening for radars) when you pick a new channel.
#4 means you have 10 seconds to move *everyone* in that "cell" off the channel.
Therefore, its going to be (ahem) "quite difficult" to use anything but the 5.725-5.825 (U-NII3) or 5.725 - 5.850 (ISM) bands for outdoor networks.
Given the above, I really can't see the need for more than 2 miniPCI sockets.
Jim
Jack Unger wrote:
John,
I know you may disagree with what I'm about to say but using 8 radios that are physically spaced so close together is very likely going to lead to interference between the radios with consequent lost throughput.
If and only if the builder of an 8-radio system uses intelligent frequency planning (for example: Ch 1, 6, and 11 on 2.4 GHz and non-overlapping frequency selections on 5 GHz) will there be any chance of avoiding interference when the system is handling moderate-to-heavy traffic. Even then, it's going to take luck (and who can rely upon luck to secure their livelihood?) to keep the interference down and the throughput up.
My suggestion: Include advice and guidance on this topic in your documentation.
jack
John Tully wrote:
The eight port mpci to pci cards main use would be to easily use eight radios in you AP platform. This should be quite useful for multiradio APs with your choice of software.
John www.mikrotik.com
RouterBOARD 18 -- 8 port mPCI to PCI board (Extreme)
http://www.routerboard.com/rb11.html#rboard18 (the back side has four ports also - tomorrow, both sides will be shown)
Price: $120 List
John www.routerboard.com www.mikrotik.com
