Design constraints for an active sensing system Insights from the Electric Sense

1
Design constraints for an active sensing
systemInsights from the Electric Sense

• Mark E. Nelson
• Beckman Institute
• Univ. of Illinois, Urbana-Champaign

2
TALK OUTLINE

• Brief background on active electrolocation
• Constraints on
• Electric field generation power considerations
• Detecting weak fields thermal noise limits
• Signal processing under low SNR conditions
• Role of multiple topographic maps?
• Coupling of sensing and action
• Summary

3
Distribution of Electric Fish
4
Black ghost knifefish (Apteronotus albifrons)
5
Electroreceptor distribution 14,000 tuberous
electroreceptor organs
mechano
MacIver, from Carr et al., 1982
6
Ecology Ethology of A. albifrons

• inhabits tropical freshwater rivers and streams
  in South America
• nocturnal hunts at night for aquatic insect
  larvae and small crustaceans
• uses electric sense for prey detection,
  navigation, social interactions

7
Self-generated Electric Field
8
Electric Organ Discharge (EOD)
9
Principle of active electrolocation
10
Electric Field GenerationPower Considerations

• Whats the metabolic cost of active sensing?
• Range related to field strength E
• Field strength falls as d-3 (inverse cube)
• Power in the electric field scales as E2
• Increasing range is expensive
• Doubling range requires 8-fold increase in
  E64-fold increase in power

11
Electric Field GenerationPower Considerations

• Weakly electric fish devote about 1 of basal
  metabolic rate to EOD production
• Pulse fish
• discharge intermittently
• higher power per EOD pulse
• lower duty cycle
• Wave fish
• discharge continuously
• lower power per EOD cycle
• 100 duty cycle

12
Electric Field GenerationPower Considerations
Long, thin tails
Short, thick tails
13
Electric Field GenerationElectric Organ Design
14
Electric Field GenerationImpedance matching
Hopkins 99
15
Principle of active electrolocation
16
Prey-capture Behavior
Daphnia magna (water flea)
1 mm
17
Prey capture behavior
18
Prey capture kinematics
Longitudinal velocity
acceleration
Distance to closest point on body surface
19
Performance constraints

• Minimum sensory range to be useful?
• Analogy driving in the fog
• Minimum useful range stopping distance
• Stopping distance velocity stopping time
• fish cruising velocity 10 cm/sec
• Stopping time reaction deceleration
• sensorimotor delay (150 msec)
• deceleration to zero (150 msec)
• Stopping distance 3 cm

20
Estimating signal strength

• Voltage perturbation at skin Df

prey volume
electrical contrast
fish E-field at prey
distance from prey to receptor
THIS FORMULA CAN BE USED TO COMPUTE THE SIGNAL AT
EVERY POINT ON THE BODY SURFACE
21
Reconstructed Electrosensory Image
22
(No Transcript)
23
Daphnia signal characteristics

• Fish can detect small prey at a distance of r 3
  cm
• Voltage perturbation at that distance is Df 1
  mV

24
Electroreceptor Constraints

• Detection of microvolt perturbations?
• Thermal noise limits

Johnson noise
effective bandwidth
10 mm cell
RMS variation in membrane potential due to
thermal fluctuations. Weaver Astumian,
Science, 1990
25
Electroreceptor constraints

• Signal 1 mV, thermal noise 30 mV
• How to improve SNR
• Multiple receptor cells per receptor organ
• (N 16, 30 mV /?16 8 mV RMS)

26
Electroreceptor Design
27
Electroreceptor constraints

• Signal 1 mV, thermal noise 30 mV
• How to improve SNR
• Multiple receptor cells per receptor organ
• Reduce bandwidth Df

receptor threshold
frequency
28
Neural coding (Probability code)
29
Change-point detectionin P-type afferent spike
trains
Phead 0.337
Phead 0.333
Phead 0.333
00010101100101010011001010000101001010
30
Signals, noise, and detectability
Extra signal spikes
Count window
31
Afferent spike train regularization
Variance-to-mean ratio F(Ik) for P-type afferents
Shuffled data(no correlations)
P-type afferents exhibit remarkable regularity on
time scales of about 50 ISIs ( 200 msec)
Ratnam Nelson J. Neurosci. 2000
32
Decreased spike train variabilityenhances signal
detectability
33
Information coding properties
34
Spike train regularization enhances
informationtransmission
Chacron et al. 2001
35
Other noise - SNR constraints

• Signal is on the order of 1 mV
• Intrinsic sensor noise (after spike train
  regularization) 1 mV
• How strong is the other background noise?
• Reafferent noise 100 mV
• Environmental noise 100 mV
• Solutions
• Subtraction of sensory expectation
• (Task-dependent) spatiotemporal filtering

36
Central Processing in the ELL
37
Design constraints for active sensing

• Upper bound on source power
• (optimize power delivery to the environment)
• Lower bound on receptor sensitivity
• (e.g., thermal noise limits)
• SNR constraints clever solutions
• (e.g., limit receptor bandwidth, spike train
  statistics, subtraction of sensory expectation,
  task-dependent spatiotemporal filtering)
• (
• Motor strategies for optimizing sensory
  acquisition
• Matching between sensory and locomotor volumes