Wireless IC Node withCompression Heuristics 

(W.I.N.C.H.) Final Design Review

Steve Jocke, Kyle Ringgenberg, Stuart Wooters<scj4s, kringg, wooters>@virginia.edu

System Data Flow Overview

avgcycleE

PIC Overview• PIC16C5X.UVa Stats

– 4 Interrupts / 3 Input ports / 5 output ports / 32 Instruction – Operating voltage of 300mV– Input clock frequency of 800KHz

• Clock divided down to 200Khz for PIC internal cycles• Approximately 200 Instructions (I-cycles) between ADC Samples• System overhead requires approximately 30 I-cycles

– Required to track input values for gain adjustment to ADC– Leaves ~ 170 I-Cycles free for compression, second core

not required

Run Length Encoding

• Replace Repetition with Occurrence Counts– Lossless: Only Identical Values Replaced

• 0 0 0 1 3 5 5 6 [3]0 1 3 [2]5 6

– Lossey: Values with a Tolerance Replaced• 0 0 0 1 3 5 5 6 [4]0 3 [3]5

– Variable Parameters• Codeword Size – Max Repetition Length• Block Size – Insert Repetition Code Every Block

– CODE x BLOCK = WORD• Tolerance – “Stray-ability”

Run Length Encoding

Run Length Encoding

Run Length Encoding

• All Variants Require Additional Repetition Blocks

• Compression Ratio– Lossless

• Poor – Worst Case is Worse than Raw– Lossey

• Excellent – Saturates to Best Compression Ratio

• Executable in O(n)– 36 Clock Cycles– 7 Registers

Delta Encoding

• Replace Values with Changes in Values– Lossless Global: Greatest Δ Dictates Remapping

• 0 0 0 1 3 5 5 6 8x 3-bit “Words” 24 bits• 0 0 1 2 2 2 1 7x 2-bit “Words” 14 bits

– Lossless Local: Each Δ Treated Independently• 0 0 1 2 2 2 1 4x 1-bit, 3x 2-bit “Word” 10 bits• Additional Data for “Word” Lengths

– Lossey Global: Mean Δ Dictates Remapping• 0 0 1 2 2 2 1 8x 1-bit 8 bits• Decoded as: 0 0 0 1 2 3 4 5

Delta Encoding

Tightly Clustered

Long Tail

Delta Encoding

Longer Tail, but no Bit Change!

0

0.5

1

1.5

2

2.5

3

ecg1 ecg2 ecg3

Raw

deltaG

deltaG'

deltaL

Delta Encoding

Delta Encoding

• All Variants Require Additional Sign Bit

• Compression Ratio– Lossless

• Mediocre – Significant Degree of Overhead Bits– Lossey

• Mediocre – Fidelity is Data Dependent

• Execution in O(n)– 37 Clock Cycles– 6 Registers

Compression Conclusions

0

2

4

6

8

10

12

14

ecg1 ecg2 ecg3

Raw

Lossless

Lossey

10x

Mixer Design

Vdd 1.2V• Active Power:

– 50GHz LO

– 100Mbs

– 9.23mW

• Active Power:– 1.5 GHz LO

– 100Mbps

– 716.67uW

15GHz Local Oscillator

• Startup Delay:– 101.626ps

• Startup Power:– 4.471mW

• Run Power:– 6.050mW

• Off Power:– 69.3914μW

LO – 3 Stage Inverter and Buffer 1.5 GHz• Startup Time:

– 2.3ns

• Startup Energy:– 6.92pJ

• Startup Power:– 3mW

• Avg Power:– 3.502mW

• Off Power:– 1.804μW

High Speed Power Examples

• VCO Core: 1mW1 51GHz• Mixer:97mW2 9-50GHz Gilbert

Cell• Amplifier 54mW3 60GHz• 1.52nJ/bit (100Mbps)• 3.04μJ/2Kbit (100Mbps)

1. Tiebout, M.; Wohlmuth, H.-D.; Simburger, W., "A 1V 51GHz fully-integrated VCO in 0.12/spl mu/m CMOS," Solid-State Circuits Conference, 2002. Digest of Technical Papers. ISSCC. 2002 IEEE International , vol.2, no., pp.238-239, 2002 2. Lin, C.-S.; Wu, P.-S.; Chang, H.-Y.; Wang, H., "A 9-50-GHz Gilbert-cell down-conversion mixer in 0.13-/spl mu/m CMOS technology," Microwave and Wireless Components Letters, IEEE , vol.16, no.5, pp. 293-295, May 2006 3. Doan, C.H.; Emami, S.; Niknejad, A.M.; Brodersen, R.W., "Millimeter-wave CMOS design," Solid-State Circuits, IEEE Journal of , vol.40, no.1, pp. 144-155, Jan. 2005

Power Consumption of PIC (No Compression)

Time μ sec

Power Consumption of PIC(cont’d)• Measurements

– Ultrasim Simulator used to measure current• Set in MS (Mixed Signal / Analog Mode)

• PIC Power Summary– Without compression

• 1.) NOP Inst Power 318.1nW (Averaged over 4 NOP’s)• 2.) AVG Inst Power 322.95nW (All runtime - 2mS)

– With compression• 3.) AVG Inst Power 336.78nW (Over Compression Algorithm)• 4.) AVG Inst Power 324.46nW (all runtime – 2mS)

– Final Stats• Overall Inst Power 323.7nW• Delta Power Modes 1.51nW

• Conclusion– Power Consumption is about the same no matter what operation

• Using free I-cycles recovers wasted power• Also could throttle back clock frequency

Energy Totals for Communication Channel

• 15GHz– On Energy (Mixer + LO):

• 152.8pJ/bit• 305.6nJ/2Kbit

– Off Power (LO):• 69.39μW

• 1.5GHz– On Energy (Mixer + LO):

• 35.06pJ/bit• 70.1nJ/2Kbit

– Off Power (LO):• 1.804μW

• Bluetooth4

– On Energy:• 297.5pJ/bit – 165nJ/bit

– Sleep Power• 825μW

– Idle Power• 8.25mW

4. Roving Networks RN-41 V1.5 11/14/07 Documentation

Summary

• Base Case (Bluetooth)

• Compression (10/1)– .324nJ/Sample (PIC)– 2.38nJ/byte (Bluetooth)– 827nJ/Sample

• No Compression– .324nJ/Sample (PIC)– 2.38nJ/byte (Bluetooth)– 842.4nJ/Sample

• Wireless• Compression

(10/1)– .324nJ/Sample (PIC)

– 1.224nJ/byte (LO)

– 74.5nJ/Sample

• No Compression– .324nJ/Sample (PIC)

– 1.224nJ/byte (LO)

– 68.8μJ/Sample