Speed%20up%20an%20FSM%20Via%20Nonlinear%20Look-ahead%20Transformation%20

1
Speed up an FSM Via Nonlinear Look-ahead
Transformation Its Application to Huffman
Decoding

• Fei Li
• Jinjun Xiong
• University of Wisconsin-Madison

2
Motivation

• Finite state machine (FSM)
• Most commonly used in sequential circuits
• Perform non-linear recursive algorithms
• Processing rate limited by iteration bound
• Applications require high-throughput
• Real-time processing
• Huffman decoding
• Nonlinear algorithm transformation techniques
  and concurrent architectures may be of help

time to compute the next state
3
Our Approach

• Nonlinear look-ahead transformation
• More state transitions in one computational step
• A formal transformation method using state
  transition matrix
• Concurrent block processing
• High-order look-ahead transformation
• High hardware complexity
• Long critical path
• Block processing
• Explore more concurrency
• Easy to pipeline

4
Example AMI Encoder

• Alternate Mask Inversion (AMI) Encoder
• map a binary sequence into a bipolar pulse
  sequence
• Maintain zero DC bias
• Formal representation of state transition
• S(n1) T(n) S(n)
• S(nm) T(nm-1) T(n) S(n)
• State indicator vector
• State transformation matrix T(n)
• Independent of current state
• Easy to perform look-ahead

5
Example AMI Encoder (cont)

• Look-ahead by three, assuming input string 010
• State Transition Matrix
• Current State 0 Current
  State 1
• Concurrent architecture w/ block processing

6
Achieved and Expected Results

• Achieved Results
• Derived formal representation of nonlinear
  look-ahead using state transition matrix
• Designed concurrent block processing architecture
• Method applied and verified on the AMI encoder
  design in Verilog
• Expected Results
• Apply the method to Huffman decoder
• Handle the issue of variable length code
• Implement it in Verilog
• Achieve high-throughput