---
license: mit
tags:
- pytorch
- safetensors
- threshold-logic
- neuromorphic
- encoding
---

# threshold-onehot-decoder

4-to-2 one-hot decoder. Converts a 4-bit one-hot representation to a 2-bit binary value.

## Function

onehot_decode(y3, y2, y1, y0) -> (a1, a0)

Input must have exactly one bit set.

## Truth Table

| y3 | y2 | y1 | y0 | a1 | a0 | Value |
|----|----|----|----|----|-----|-------|
| 0 | 0 | 0 | 1 | 0 | 0 | 0 |
| 0 | 0 | 1 | 0 | 0 | 1 | 1 |
| 0 | 1 | 0 | 0 | 1 | 0 | 2 |
| 1 | 0 | 0 | 0 | 1 | 1 | 3 |

## Architecture

Single-layer implementation:

```
  y3   y2   y1   y0
   │    │    │    │
   └────┴────┴────┘
         │
    ┌────┴────┐
    │         │
    ▼         ▼
  ┌───┐     ┌───┐
  │a1 │     │a0 │  Layer 1
  │OR │     │OR │
  └───┘     └───┘
    │         │
    ▼         ▼
```

Each output is a single OR gate:
- a0 = OR(y1, y3): w=[1,0,1,0], b=-1
- a1 = OR(y2, y3): w=[1,1,0,0], b=-1

The decoder recognizes that:
- Bit 0 of position = 1 for positions 1 and 3
- Bit 1 of position = 1 for positions 2 and 3

## Parameters

| | |
|---|---|
| Inputs | 4 |
| Outputs | 2 |
| Neurons | 2 |
| Layers | 1 |
| Parameters | 10 |
| Magnitude | 6 |

## Usage

```python
from safetensors.torch import load_file
import torch

w = load_file('model.safetensors')

def decode(y3, y2, y1, y0):
    inp = torch.tensor([float(y3), float(y2), float(y1), float(y0)])
    a0 = int((inp @ w['a0.weight'].T + w['a0.bias'] >= 0).item())
    a1 = int((inp @ w['a1.weight'].T + w['a1.bias'] >= 0).item())
    return a1, a0

# decode(0, 1, 0, 0) = (1, 0)  # value 2
```

## Applications

- Priority encoder output processing
- State machine decoding
- Memory address reconstruction
- Neural network output interpretation

## License

MIT