Download presentation
Presentation is loading. Please wait.
1
Group M3 Craig LeVan Jacob Thomas Nick Marwaha Darren Shultz Project Manager: Zachary Menegakis February 14, 2005 MILESTONE 4 Gate Level Design DSP 'Swiss Army Knife' Overall Project Objective: General Purpose Digital Signal Processing Chip
2
STATUS Design Proposal (Done) Architecture (Done) Size Estimates/Floorplan/Verilog (66%) Gate Level Design (80% - Debugging) To Be Done Comb Reengineering Current design causes errors due to the mantissa Current prototype solution is ENORMOUS!!! Affects four of the fifteen functions of the chip Schematic Convert few remaining Verilog blocks Make comb adjustments Verification Retest for comb functions Layout (0%)
3
DESIGN DECISIONS Circuit Speed Goal Most audio signals played at ~44 kHz Motorola’s audio super chip operates at 120 MIPS (performs functions such as Dolby Digital) Use Motorola as benchmark which equals 120 MHz for us However, want to keep power as a focus as well Type of Multiplier (for mantissa within floating point multiply) Chose matrix multiply since easy to implement Others such as Wallace offer no real advantage at 6 bits, while are more difficult to layout Comparator Using carryout of subtractor (also used in previous projects) Shifting Logic Using combinational barrel shifter for adder and normalization Comb Solution Alternative B
4
COMB FILTER Assumptions about Floating point operation of x -n proved incorrect Initial re-design is unrealistic
5
DESIGN DECISIONS cont Namea0a1a2b0b1b2c1N 1Differencer100100x 2Integrator1101000x 3Leaky Integrator1101000x 4Comb Filter10010018 5Bandpass Filter10100116 6CIC Interpolation Filter11010018 7dc Bias Removal1a.b0100x 8First-Order Equalizer1a.b0 100x 9Audio Comb10a.b1000x 10Moving Averager1101/N0018 11Second-Order IIR Filter1a.bbb 0x 12First-Order Delay Network1a.bbb 10x 13Second-Order Delay Network1a.bbb 10x 14Real Oscillator12cos(x)10 xx 15Second-Order Equalizer1a.b*cos(x)a.b1a.b*cos(x)1/a.b0x The Affected Functions
6
DESIGN DECISIONS cont Alternative B (not what we want) Can remove while only eliminating 1 function, and significantly reduce transistor count
7
UPDATED DATA FLOW
8
UPDATED DATA FLOW cont
9
FLOORPLAN UPDATE
10
SIZE ESTIMATES
11
SCHEMATIC Biquad Top Level
12
SCHEMATIC cont Floating Point Multiply
13
SCHEMATIC cont Fixed Point Multiply
14
VERIFICATION VSIM 1> run # x xxxxxx xxxxx * x xxxxxx xxxxx = x xxxxxx xxxxx # 0 000000 00000 * 0 000000 00000 = 0 000000 00000 # 0 011110 00000 * 1 011101 11000 = 1 011100 11000 # 0 100001 00100 * 0 100000 01000 = 0 100010 01101 # 0 100001 01110 * 0 100000 00001 = 0 100010 01111 # 0 100001 11100 * 0 100000 11110 = 0 100011 11010 # 0 100100 11110 * 0 100010 11000 = 0 101000 10110 # 1 100100 11110 * 1 100010 11000 = 0 101000 10110 # 0 100001 00010 * 0 100001 11110 = 0 100100 00000 # ** Note: $finish : fp_mult_tb0.v(41) # Time: 9 ns Iteration: 0 Instance: /tester
15
VERIFICATION cont VSIM 1> # x xxxxxx xxxxx + x xxxxxx xxxxx = x xxxxxx xxxxx # 0 000000 00000 + 0 000000 00000 = 0 000000 00000 # 0 011110 00000 + 1 011101 11000 = 0 011011 00000 # 0 100001 00100 + 0 100000 01000 = 0 100001 11000 # 0 100001 01110 + 0 100000 00001 = 0 100001 11110 # 0 100001 11100 + 0 100000 11110 = 0 100010 01101 # 0 100100 11110 + 0 100010 11000 = 0 100101 00110 # 1 100100 11110 + 1 100010 11000 = 1 100101 00110 # 0 100001 00010 + 0 100001 11110 = 0 100010 10000 # ** Note: $finish : fp_add_tb0.v(40) # Time: 9 ns Iteration: 0 Instance: /tester
16
CRITICAL PATH When Comb is being used N = 16 (especially with alternative B) Biquad is always used There are no specific operation codes Function being performed is based on coefficient values
17
PROBLEMS & QUESTIONS Comb MUST find an Alternative A to keep size down OR should we eliminate Comb??? Efficient Comparator Is there any method better than using the carryout of a subtractor? Structural Barrel Shifter / Divider Schematic wiring
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.