RLS Adaptive Filter

Overview

The eSi-QRD-LSL is a QR-Decomposition Least Squares Lattice (QRD-LSL) core is a high throughput IP core that finds application in challenging adaptive filtering. It computes the Recursive Least Squares solution for time series data in a numerically efficient and stable way.

Key Features

More than 100x faster than a pure software implementation
Lattice implementation with joint process estimation
Low hardware usage iterating around a single lattice stage
Fully Recursive Least Squares (RLS) solution for time domain data
A Posteriori error and filtered joint process data output
Numerically stable orthogonal transforms
Run-time setting of forgetting factor
Run-time filter length setting
Run-time filter coefficient update freeze
Run-time restart adaption
Symmetric rounding and saturation
Parameterised bit-widths for minimum hardware
AXI4 streaming data interface

Benefits

Superior convergence to LMS adaptive filter
Lower mean squared error than LMS adaptive filter
Available in shared single stage or fully unrolled architecture
Shared single stage: 21 clock cycles per stage
Unrolled: 21 clock cycles per input sample
Low resource usage

Applications

Echo cancellation
Industrial processing
Communications
Radar/Sonar

Deliverables

RTL source code
Testbench
Bit exact C++ model
Synthesis scripts
FPGA and ASIC implementation

Technical Specifications

Availability

Available Now

Request Info