A Subsampling Radio Architecture for 310 GHz UWB

1
A Subsampling Radio Architecture for 3-10 GHz UWB

• Mike Shuo-Wei Chen
• Prof. Bob Brodersen
• June 13, 2003 BWRC Retreat

2
Concept of Subsampling

• Begin with a bandlimited signal at a certain
  frequency band Fl Fh

B

• Receiver samples at Fs, the resulting signal
  spectrum is the repetitive replicas of integer
  multiple fs

p
-p
3
Motivation

• Motivation
• - Among many existed wireless systems, UWB has
  to be somewhat different, such as cheaper and
  lower power.
• - Subsampling radio has fewer components
  compared to traditional heterodyne architecture.
  No analog multipliers and LO required for mixing.
  
• Challenges of Subsampling
• - All thermal noise power between -FlFl will
  fold back to signal band. The amount of aliasing
  is proportional to fc/B.
• - RF bandpass filter design is harder than IF
  or baseband.
• - Tighter sampling jitter requirement.

4
Why Promising for UWB?

• Much lower fc/B ratio than narrowband system
• Lets say 1 GHz pulse centered from 310 GHz
• Noise level is dominated by the in-band noise
  interference
• Giga-hertz bandwidth will inevitably pick up
  the signals from other wireless systems
• Q of RF bandpass filter is low
• Lower number of ADC bits

5
Integrated Analog Front End

• Radio Architecture
• - Wideband antenna
• - Wideband amplifier / matching network
• - RF bandpass filtering (low Q filter)
• - High bandwidth sample and track
• - High-speed and low resolution ADC
• - Sampling Clock generator
• - DSP

6
Different Transmitter

• Narrowband System
• A baseband signal is mixed up to carrier
  frequency

• UWB system
• Pulser drives antenna and generates a
  passband signal without mixer

7
How to recover the information?

• Reach the matched filter bound digitally without
  wideband analog processing. Another good example
  of taking the advantage of technology scaling.
• How to know the matched filter response?
• A Simply use running average during training
  sequence. The best linear estimator, if the noise
  is AWGN.
• A small shift of sampling time will change the
  sampled waveform dramatically. A timing recovery
  issue! Matched filtering failed
• A Propose a Complex signal based digital
  backend

8
Sampling Process
To
Ts
A sampling offset maps to a constant phase shift.
9
Pulse Used in Simulation

• UWB pulse measured with monopole antenna. Pulse
  is then filtered by an ideal 3-4 GHz bandpass
  filter in Matlab.

10
Sampling Offset Effects

• Sampled pulse shape changes dramatically with
  fractional of sampling spacing!

11
Analytic Signal Processing

• Convert a discrete real signal into a discrete
  analytic signal via Hilbert transformer.
• Real and imaginary part of the analytic signal
  are two orthogonal sets. The analogy is I and Q
  channel!
• Analytic MF response h(t) s(T-t).

Extract the hidden info by analytic signaling!
12
Constellation _at_ Analytic MF outputs

• Timing offset cause the constellation to rotate.

Imag
0Ts
0.05Ts
signalnoise
noise
Real
0.1Ts
0.15Ts
13
Trajectory with infinite-bit ADC

• Sampling offset varies from 0.1Ts to 4Ts.
• SNR of Real, Imag, Abs of analytic MF output.
• Timing Information!

14
Trajectory with 1-bit ADC

• More averaging is needed or the robustness to the
  sampling offset is reduced.

15
Subsampling Analytic Filtering

• Possible application
• Data communication or Ranging at a low cost

16
Conclusion

• A subsampling analog front end combined with
  analytic signal processing has been proposed for
  passband UWB communication.
• By the magnitude of complex MF output, timing
  recovery has been done without interpolation or
  oversampling. Thus, it makes digital matched
  filtering promising and practical.
• This technique also achieves a very fine timing
  resolution smaller than sampling rate, which
  implies high accuracy ranging capability. This is
  a special feature of UWB.