Electrosensory data acquisition and signal processing strategies in electric fish Mark E. Nelson Beckman Institute Univ. of Illinois, Urbana-Champaign
How Electric Fish Work
Distribution of Electric Fish Fish tank upstairs black ghost knifefish elephant- nose fish
Electric Organ Discharge (EOD) - Spatial
EOD - Temporal
Electric Organ Discharge (EOD)
Principle of active electrolocation
mechano MacIver, from Carr et al., 1982 Electroreceptors ~15,000 tuberous electroreceptor organs 1 nerve fiber per electroreceptor organ up to 1000 spikes/s per nerve fiber
Individual Sensors (Electroreceptors) V IN nerve spikes OUT
Neural coding in electrosensory afferent fibers
Probability coding (P-type) afferent spike trains P head = P head = P head = 0.333
Principle of active electrolocation
Electrosensory Image Formation
Prey-capture video analysis
Prey capture behavior
Fish Body Model
Motion capture software
MOVIE: prey capture behavior
Electrosensory Image Reconstruction
Voltage perturbation at skin : Estimating Daphnia signal strength electrical contrast prey volume 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
MOVIE: Electrosensory Images
System Capabilities Electric fish can analyze electrosensory images to extract information on target direction (bearing) distance size shape composition (impedance)
Distance Discrimination
Shape Discrimination
Shape Generalization
Shape “completion”
Impedance Discrimination
How Do They Do It? Electric fish analyze dynamic 2D electrosensory images on the body surface to determine target direction, distance, size, shape and composition (impedance) Fish might perform an inverse mapping from 2D sensor data to obtain a dense 3D neural representation of world conductivity sensor data 3D conductivity action Alternatively, fish might use sensor data to directly estimate target parameters sensor data target parameters action
Parameter estimation (bearing)
Parameter Estimation (cont.)
Dynamic Movement Strategies Fish are constantly in motion not a single, static ‘snapshot’ dynamic, spatiotemporal data stream With respect to target objects in the environment, fish body movements simultaneously influence the relative positioning of the sensor array the electric organ effector organs (e.g. mouth)
MOVIE: Electrosensory Images
Active motor strategies: Dorsal roll toward prey
Probing Motor Acts chin probing back-and-forth (va et vient ) lateral probing tangential probing stationary probing
Fish exploring a 4 cm cube
CNS Signal Processing Strategies Multi-scale filtering spatial and temporal Adaptive background subtraction tail-bend suppression Attentional ‘spotlight’ mechanisms local gain control
Multiple Maps
Multi-scale Filtering INPUT (from skin receptors) Centromedial map High spatial acuity Low temporal acuity Centrolateral map Inter spatial acuity Inter temporal acuity Lateral map Low spatial acuity High temporal acuity temporal integration both spatial integration HINDBRAIN PROCESSING PERIPHERAL SENSORS
Adaptive Background Subtraction
Attentional ‘spotlight’ mechanism
Summary Fish can evaluate direction, distance, size, shape and composition of target objects How? model-based parameter estimation based on 2D image analysis, not full 3D reconstruction presumably some sort of (adaptive) (extended) (unscented) Kalman-like algorithm extensive pre-filtering (virtual sensors?) self-calibrating, adaptive noise suppression, multi- scale spatial and temporal signal averaging dynamic control of source and array position
Acknowledgements Colleagues Curtis Bell (OHSU) Len Maler (Univ. Ottawa) Gerhard von der Emde (Univ. Bonn) Nelson Lab Members Ling Chen, Rüdiger Krahe, Malcolm MacIver Funding Agencies NIMH, NSF