Capacity Planning Discussion
Noise is a factor in the frequency domain (C/I) and the time domain (Eb/No- minimized by multipath)
Signal to Noise Ratio can be improved by using multiple antennas (simple diversity), maximum ratio combining, guard band between bits (to negate the destructive effect of ISI), space time coding, Spacial multiplexing and beam forming.
Since Wimax uses all of these techniques it is very difficult to determine what state any particular carrier is in at any particular time. Adaptive Modulation and Coding along with Adaptive Mode MIMO (switches between matrix A-Space Time Block Coding and Matrix B-Spacial Multiplexing) provide a system in a constant state of change.
The amount of multipath at each CPE cannot be predicted.
AAS shows the best ability (theoretically) of improving C/I but negates the Inter Symbol Interference gains (ISI) of MIMO and guard bands (time).
Capacity considerations are: (using John Little's Law of Queueing)
Time in System = Waiting Time + Service Time
Number in System = Number Waiting + Number Being Served
Arrival Rate = Number Waiting / Waiting Time
Number being Served = Arrival Rate x Service Time
Number Waiting = Arrival rate x Time Waiting
Delay Probability = Link Utilization
etc, etc, etc..... It's a lot more difficult that using an Erlang B chart for voice.
I highly recommend attending the Wimax RF Designer Certification course offered by the Wimax University if you are going to design and operate a Wimax network.
A really good Wimax RF planning tool will be able to calculate capacity, queueing, and take into account the effects of Adaptive Modulation and Coding (BAND AMC), Adaptive MIMO (switches from Matrix A to Matrix B) and closed loop MIMO (Beam Forming).
I look forward to your comments......
Coming from the TDM voice world to flat IP networks, it was unusual to transition to a paradigm of soft-capacity (CDMA has soft capacity, but blocks calls per Erlang B). The more load you put on an IP network, simply the longer the packet delay. There's no blocking, the load is just absorbed (unless you run out of buffer, which is a huge problem because it doesn't get rid of the excess load, TCP just resends it on top of the other data already trying to be sent, exacerbating the problem). This why call blocking (and QOS) implemented on a VoIP network is so important (via SIP or proprietary means), because if the load is too big, latency and jitter become intolerable, and ALL calls get screwed.
Radio Resource Management systems are expected to be the most difficult for vendors to develop (and biggest area of vendor differentiation). The capacity of the system will be heavily dependent on these algorithms. It has to marry pure capacity schemes (preference to high SNR) with fairness (low SNR equally needing BW); throw in inter-cellular subcarrier sharing (fractional reuse), 5ms traffic decisions ,and QOS, and it becomes a monster traffic engineering problem. I've read of weighting algorithms whereby the probability of getting resources is dependent on SNR and how long you've been stuck in the Queue. What they try to do is wait and see if you get a better SNR later on, so high SNRs get privileged in the short term (unfair, but higher capacity vis-à-vis modulation/FEC state), but equal fairness in the long-term. Fun, fun….
Some capacity-related docs attached...
Attachments
-
ChannelCapacity.pdf, 253 KB - CDC04.pdf, 119 KB
- InOWo2004.pdf, 107 KB
Here is a chart to help visualize the possibilities.
Attachments
OFDMA rates.jpg
[get this widget]
