Part 3: Channel Capacity ECEN478 Shuguang Cui http://www.eecs.berkeley.edu/~dtse/main.pdf
Shannon Capacity Defined as the maximum mutual information across channel (need some background reading) Maximum error-free data rate a channel can support. Theoretical limit (usually don’t know how to achieve) Inherent channel characteristics Under system resource constraints We focus on AWGN channel with fading
AWGN Channel Capacity Goldsmith, Figure 4.1 AWGN channel capacity, bandwidth W (or B), deterministic gain: g[i]=1 is known and fixed Per dimension: Bits/s/Hz 0.5 Total: Bits/s If average received power is watts and single-sided noise PSD is watts/Hz,
Power and Bandwidth Limited Regimes Bandwidth limited regime capacity logarithmic in power, approximately linear in bandwidth. Power limited regime capacity linear in power, insensitive to bandwidth. If B goes to infinity?
Capacity Curve
Shannon Limit in AWGN channel What is the minimum SNR per bit (Eb/N0) for reliable communications? Where: for small
Capacity of Flat-Fading Channels Capacity defines theoretical rate limit Maximum error free rate a channel can support Depends on what is known about channel CSI: channel state information Unknown fading: Worst-case channel capacity Only fading statistics known Hard to find capacity
Capacity of fast fading channel : Flat Rayleigh, receiver knows. Unit BW, B=1. Fast fading, with a certain decoding delay requirement, we can transmit time duration LTc (L>>1), i.e., L coherence time periods. For l-th coherence time period, we have roughly the same gain: The received SNR: The capacity (Rx knows CSI): Average capacity over L period:
Fast fading, only Rx knows CSI As L goes large: This is so called Ergodic Capacity. Achievable even only receiver knows the channel state. Less than AWGN
Example Fading with two states Ergodic capacity AWGN counterpart
Fading Known at both Transmitter and Receiver For fixed transmit power, same as only receiver knowledge of fading, but easy to implement Transmit power can also be adapted Leads to optimization problem:
An equivalent approach: power allocation over time Channel model: Notation: Subject to:
Optimal solution Use Lagrangian multiplier method, we have the water-filling solution: To define the water level, solve:
Asymptotic results As L goes to infinity, we have: The solution converges to be the same as the textbook approach!
Example Fading with two states Water-filling Where is the water level? Three possible cases for
Water-filling over time
Implementation with discrete states Goldsmith, Fig 4.4 We only need N sets of optimal AWGN codebooks. (We need feedback channel to know the channel state.)
Performance Comparison At high SNR, waterfilling does not provide any gain. Transmitter knowledge allows rate adaptation and simplifies coding.
Time Invariant Frequency Selective Channel We have multiple parallel AWGN channels with a sum power constraint! Yes, water-filling!
Multicarrier system in ISI channel
OFDM-discrete implementation of multi-carrier system Transmitter
OFDM receiver FFT matrix:
Time Varying Frequency Selective Channel Maximize: s. t.: Two-dimension Water-filling!
Summary of Single User Capacity Fast fading channel: Ergodic capacity: achievable with one fading code or multiple sets of AWGN codes Power allocation is WF over distribution Frequency selective fast fading channel: Ergodic capacity is achieved with 2-D WF