Ontological Layer: Signal Interaction
Digital Convolution
The Sliding Engine of Discrete Processing
The Mechanics of Overlap
How does a digital filter actually process an input sequence? It does not happen instantly. The system's impulse response is reversed, and then mathematically slid across the input data, overlapping sample by sample. At every shift, the overlapping values are multiplied and summed to produce a single point of output. This is Linear Convolution. But because this raw sliding calculation is computationally brutal, we use Zero-Padding and the FFT from Realm 10 to cheat the math, performing high-speed Fast Convolution via multiplication in the frequency domain.
The 14 Convolution Nodes
Ontological Mapping: From Graphical Sliding to Fast Algorithms
What is Convolution?
- Mathematical combination of two sequences
- Calculates LTI system output
- Weighted accumulation of past inputs
- Sliding overlap mechanism
Discrete Formula
- y[n] = Σ x[k]h[n-k]
- k as the dummy integration index
- Shifted sequence interpretation
- Reversal (h[-k]) and Shift (n)
Graphical Method
- Step 1: Write x[k] and h[k]
- Step 2: Flip h[k] to h[-k]
- Step 3: Shift by n
- Step 4: Multiply overlapping points & Sum
Support & Length
- Finite vs Infinite support
- No overlap → y[n]=0
- Partial vs Full overlap intervals
- Output Length = L + M - 1
Linear Convolution
- Standard discrete convolution
- Sequence boundaries naturally expand
- Zero values assumed outside support
- Algebraic and Tabular computation methods
System Properties
- Commutative: x*h = h*x (Swap order)
- Associative: Cascaded systems (x*h₁)*h₂
- Distributive: Parallel systems x*(h₁+h₂)
- Shift Property propagates delays
LTI Connection
- Decomposing input into shifted impulses
- Superposition of impulse responses
- Why h[n] completely characterizes system
- Bridge to Realm 02 concepts
Elementary Convolutions
- Convolution with unit impulse δ[n]
- Identity operation: x[n]*δ[n] = x[n]
- Shifted impulse: x[n]*δ[n-k] = x[n-k]
- Convolution with unit step (Accumulation)
Circular Convolution
- Required for DFT/FFT computation
- Periodicity assumption of sequences
- Modulo indexing (wraparound behavior)
- y[n] = Σ x[k]h[(n-k) mod N]
Linear vs Circular
- Linear length (L+M-1) vs Circular length (N)
- Circular causes time-domain aliasing (wrap)
- Linear models physical systems
- Circular is a mathematical FFT artifact
Zero Padding
- Appending zeros to prevent wraparound
- Extending sequence lengths to N
- Condition: N ≥ L + M - 1
- Forces Circular result to equal Linear result
Fast Convolution
- Time Convolution ↔ Freq Multiplication
- y[n] = IDFT( DFT(x) · DFT(h) )
- Requires proper Zero-Padding
- Reduces O(N²) cost to O(N log N) via FFT
Convolution Engine
- Method: Graphical vs Algebraic vs Tabular
- Type: Linear vs Circular
- Algorithm: Direct Computation vs FFT-based
- Block processing (Overlap-Add/Save) preview
14 — The Sliding Laboratory
Digital Convolution is the fundamental operation that physically implements digital filters. By mastering how one sequence slides through another, and how to accelerate this process using the FFT from Realm 10, we unlock true real-time signal processing. This prepares us for the final analytical tools in Realm 12: Signal Operations & Analysis.