| name | decorrelation-grid-cell-distance |
| description | Distance coding via de-correlation of heterogeneous grid cell populations. Mathematical theory showing how small variability in grid properties enables distance encoding through population activity de-correlation, with non-intuitive 'sweet spot' predictions and range-distinguishability trade-offs. Activation: grid cells, distance coding, de-correlation, medial entorhinal cortex, navigation, population coding, heterogeneity, spatial navigation, place cells, path integration |
| metadata | {"arxiv_id":"2511.08292","published":"2025-11-11","authors":"Pritipriya Dasbehera, Akshunna S. Dogra, William T. Redman","tags":["grid-cells","distance-coding","de-correlation","MEC","navigation","population-coding","heterogeneity"]} |
Distance by De-correlation: Computing Distance with Heterogeneous Grid Cells
Overview
Mathematical theory for how grid cells in medial entorhinal cortex (MEC) encode distance through de-correlation of population activity, exploiting small but robust heterogeneity in grid properties.
Core Theory
Key Insight
- Grid cell populations have small but robust heterogeneity in grid spacing, orientation, and phase
- Distance between locations can be decoded from the de-correlation of population activity patterns
- This is NOT rate coding — it's a population-level statistical computation
Mathematical Framework (1D)
- Population activity at position x: vector of grid cell firing rates
- De-correlation function: C(d) = correlation between activity patterns separated by distance d
- C(d) decreases monotonically for small d, then oscillates with grid period
- Distance estimate: invert the de-correlation function
Non-intuitive Predictions
- Sweet spot: Some further distances are better encoded than some nearer distances
- Range-distinguishability trade-off: More variable grid properties → wider encoding range but lower precision
- Optimal heterogeneity: Measured grid cell variability (~5-10% CV) strikes a balance enabling encoding up to several meters
Extension to 2D
- 2D de-correlation depends on both distance AND direction
- Anisotropic encoding: different precision in different directions
- Grid property variability controls the isotropy of distance encoding
Methodology
De-correlation-Based Distance Decoder
- Record population activity at start and end positions
- Compute correlation coefficient between activity vectors
- Map correlation → distance using calibrated de-correlation function
- Decoder performance validated against rodent behavioral data
Simulation Framework
- Noisy grid cells with realistic firing field variability
- Heterogeneous grid spacing (Gaussian distribution, CV ~5-10%)
- Population sizes: 50-500 cells
- Test distances: 0.1m to 10m
Key Results
- Decoder achieves ~15% error for distances 0.5-3m (consistent with rodent behavioral data)
- Sweet spot at ~1-2m matches published rodent distance estimation experiments
- Trade-off curve: heterogeneity CV of 5-10% is near-optimal for encoding 0.5-5m range
Biological Plausibility
- Requires only local computation (correlation of population vectors)
- Compatible with known MEC circuit architecture
- Explains why grid cells have heterogeneity (previously considered noise)
- Predicts that distance-encoding neurons should exist in MEC/subiculum
Pitfalls
- Theory assumes stationary grid fields (no remapping during navigation)
- Does not account for velocity modulation of grid cell firing
- 1D theory may not fully capture 2D navigation complexity
- Requires sufficient population size (>50 cells) for reliable decoding
Related Concepts
- Grid cells (Hafting et al., 2005)
- Path integration
- Continuous attractor networks
- Population vector decoding
- Place cells (hippocampus)
- Head direction cells