A Low Power Asynchronous GPS Baseband Processor

Benjamin Z. Tang, Stephen Longfield, Jr., Sunil A. Bhave, Rajit Manohar

Cornell University

05/07/2012 - 1/18 Benjamin Tang

Motivation

1980s 1990s 2000s 2010s FUTURE

Need

continuous

operation, much

lower power Decreasing power, but still too high

Augmented reality

Micro robotics

navigation

Location-based services

05/07/2012 - 2/18 Benjamin Tang

How Does GPS Work? • GPS L1 civil signal

L1 carrier Pseudorandom noise code

(PRN) • 1ms repeat period, 1.023MHz • Unique for each satellite

Navigation data

τ1

τ3

05/07/2012 - 3/18

τ2

Navigation data

PRN code

L1 carrier

GPS satellite

transmitted

signal

Receiver

Satellite

Benjamin Tang

How Does Receiver Know… • Which satellite’s signal was received?

Use CDMA

• Where the satellite is? Orbital information in navigation data

• When was the signal transmitted? Navigation data + PRN code phase

Navigation data

PRN code

L1 carrier

GPS satellite

transmitted

signal

05/07/2012 - 4/18 Benjamin Tang

GPS Receiver

05/07/2012 - 5/18

“Channel”

Medium

power

<10mW

Negligible

power More power-hungry

~20-100mW

GPS RF

FrontendDigital Samples

GPS Baseband

ProcessingGPS Baseband

ProcessingGPS Baseband

ProcessingGPS Baseband

ProcessingGPS Baseband

ProcessingGPS Baseband

Processing

Measurements & Decoded

MessagePosition

calculation

Our focus

Benjamin Tang

GPS Baseband Processing

05/07/2012 - 6/18

• Correlation in CDMA

Generate signal replica

Multiply and accumulate

Digital Samples

Tracking

Data

Decode

Controls

Signal Replica

Accumulators

Measurements & Decoded

Message

~6 MHz ,

kHz

1kHz –

50Hz

Received

Receiver-generated

code replica

“Channel” “Correlators”

Output once every 1ms

~1.023MHz

&

Subsystems

should run at

their natural

frequencies to be

power-efficient

Benjamin Tang

Baseband Processor Design Options

Options Correlators Tracking Decode

Option 1 Software Software Software

Option 2 Hardware Software Software

Option 3 Hardware Hardware Hardware

Typical synchronous design issue: What clocks to use?

• Shared with front end oscillator crystal Optimized for one particular front end

Clock ratios, unnecessary power

• Independent oscillator crystals Optimizations less front end dependent

Clock ratios

Processor clock >> sampling clock, unnecessary power

Our implementation

Asynchronous:

Each subsystem

only runs as fast

as it needs to

05/07/2012 - 7/18 Benjamin Tang

Asynchronous GPS Baseband Processor

• 6 channels

• Selected optimizations

• QDI and bundled data

05/07/2012 - 8/18

Digital Samples

Tracking

Data

Decode

Controls

Signal Replica

Accumulators

Measurements & Decoded

Message

Tracking

Digital Samples

Data

Decode

Controls

Signal Replica

Accumulators

Measurements & Decoded

Message

Buffer

Shared tracking loops

Asymmetric

acquisition

QDI

Bundled-data

Benjamin Tang

Asymmetric Acquisition Full Acquisition (Other receivers) Asymmetric Acquisition (Our receiver)

(+) Acquires: satellite ID, code phase offset and Doppler frequency

(-) Acquires: code phase offset, the rest from software

(-) FFT engine and memory or thousands of correlators

(+) Use pre-existing correlators

Full acquisition not needed often. Use asymmetric acquisition scheme.

05/07/2012 - 9/18

Digital Samples

Tracking

Data

Decode

Measurements & Decoded

Message

Controls

Signal Replica

Accumulators

Reduced hardware,

Reduced area,

Reduced power

Benjamin Tang

Accumulators

• Operate at input frequency • 6 accumulators per channel • 3-bit inputs iteratively added to 16-bit sum • Only dump output once every 1ms • Higher order bits do not switch often

05/07/2012 - 10/18

3

IN

Reg

OUT

16

16DUMP

16

DUMP

Bit=1

Bit=0 Bit=0

Benjamin Tang

Accumulators

• Standard 3-bit accumulator coupled with a 13-bit constant time counter

• Concatenate results at DUMP

• Naïve 16-bit accumulator: ~40μW Counter-based accumulator: ~10 μW

05/07/2012 - 11/18

4X

less

power

3

IN

Reg

OUT

16

16DUMP

16

3

IN

Reg

3

3

Counter

3

Carry

out{b,a}

a

b

13

OUT

16MSB

DUMP

Benjamin Tang

Tracking Loops • Tightly coupled feedback loops

Need to provide updates before the next data sample

Fast tracking loops, power hungry

Digital Samples

Tracking

Data

Decode

Controls

Signal Replica

Accumulators

Measurements & Decoded

Message

Tracking

Digital Samples

Data

Decode

Controls

Signal Replica

Accumulators

Measurements & Decoded

Message

05/07/2012 - 12/18

Buffer

• Defer updates Slow tracking loops, shared between all channels, saves power

Benjamin Tang

Tracking Loops

• Frequency Locked Loop (FLL), Phase Locked Loop (PLL) and Delay Locked Loop (DLL)

• Computations involve vector magnitude, arctangent, multiplication and division operations. Simplify:

Fixed point arithmetic, bundled-data

Apply Taylor series small angle approximation:

Apply modified version of Robertson approximation:

1tan

2 2 1 14 4

max ,A I Q I Q Q I

05/07/2012 - 13/18

Position error increases by ~1m

on average

Benjamin Tang

Receiver Performance Simulations

• Transistor-level implementation of our system • Position accuracy simulation

60 seconds of signal from commercial GPS signal simulator No added atmospheric, ionospheric and multipath errors

05/07/2012 - 14/18

3D-RMS error <4m

Benjamin Tang

Power Simulations

SPICE simulation: Vdd=1V, T=25oC, 90nm technology

1.4mW during

continuous tracking

05/07/2012 - 15/18

Subsystems Acquisition (μW)

(6 Channels)

Track (μW)

(6 Channels)

Correlators

Code Generator 41.8 39.9

Carrier NCO 477.4 442.8

Code NCO 439.4 400.2

Accumulators 367.3 359.9

Tracking Loops 5.5 5.8

Data Decode 1.9 2.1

Controls, Support 240.3 239.1

Total 1.49mW 1.41mW

Benjamin Tang

Comparison

• Other contemporary GPS receivers (SOCs with integrated RF front end and baseband processing)

MediaTek (J.-M. Wei, et al., ISSCC 2009)

STMicroelectronics (G. Gramegna, et al., JSSC 2006) 05/07/2012 - 16/18

Name This work MediaTek ST

Process 90nm 0.11 μm 0.18 μm

Voltage (V) 1.0 1.2 1.6

Number of Channels 6 22 12

System Power (mW) 1.4 34.0 56.0

RF Power (mW) - 19.5 20.0

Baseband Power (mW) 1.4 14.5 36.0

Baseband Power/Channel (mW) 0.2 0.7 3.0

3-D rms Error (m) 3.9 - 3.0

3X lower

power per

channel Comparable

accuracy

10X lower

power

Benjamin Tang

Conclusion

• Transistor-level implementation of a low power asynchronous GPS baseband processor

Only runs as fast as it needs to

• Selected optimizations:

Asymmetric acquisition

Counter-based accumulators

Shared bundled-data tracking loops

05/07/2012 - 17/18

1.4mW

3D-RMS < 4ms

Benjamin Tang

Acknowledgement

• Dr. Paul Kintner

• DARPA HI-MEMS

• National Science Foundation

05/07/2012 - 18/18 Benjamin Tang

A Low Power Asynchronous GPS Baseband Processor

Benjamin Z. Tang, Stephen Longfield, Jr., Sunil A. Bhave, Rajit Manohar

Cornell University

Benjamin Tang