DPD Algorithm

Digital Predistortion Algorithm

/dij-ih-tul pree-dis-tor-shun/
DPD algorithms model the PA's nonlinear behavior and generate a pre-corrected input signal that, after passing through the PA, produces a linear output. Common algorithms include memoryless polynomial (simplest), memory polynomial (captures PA memory effects), GMP (Generalized Memory Polynomial), and neural network-based (most flexible). DPD algorithms are implemented in FPGAs or DSPs and adapt in real-time using feedback from the PA output.
Category: PA Linearization
Related to: Digital Predistortion, Amplifier, ACLR, EVM, GaN
Units: dB

Understanding DPD Algorithms

DPD algorithms are the software core of modern PA linearization systems. The algorithm's accuracy in modeling the PA's nonlinear behavior determines the achievable ACLR improvement and overall linearization performance.

DPD Model Types

  • Memoryless polynomial: y = a1*x + a3*x*|x|^2 + a5*x*|x|^4. Simple, fast. Works for PAs without significant memory.
  • Memory polynomial: Includes delayed samples to capture PA memory effects. Most common practical DPD model.
  • GMP: Adds cross-terms between current and delayed samples. Better for wideband PAs with strong memory.
  • Neural network: ML-based model. Most flexible. Higher complexity but captures arbitrary nonlinear behavior.

DPD Performance

  • Memoryless: 10-15 dB ACLR improvement.
  • Memory polynomial: 15-25 dB improvement.
  • GMP: 20-30 dB improvement for wideband signals.
Common Questions

Frequently Asked Questions

What is a DPD algorithm?

DPD algorithms model PA nonlinearity and generate pre-corrected input signals. Common types: memoryless polynomial (simple), memory polynomial (standard), GMP (wideband), and neural network (most flexible). Achieves 15-30 dB ACLR improvement.

How does DPD adapt in real-time?

A feedback path samples the PA output and compares it to the desired signal. The algorithm updates its coefficients (using LMS or similar) to minimize the error. Adaptation rate: milliseconds to seconds depending on the algorithm complexity.

What limits DPD performance?

PA observation bandwidth (must be 3-5x signal bandwidth), DAC/ADC resolution (limits correction accuracy), feedback path delay (limits loop bandwidth), and PA memory depth (longer memory needs more complex models).

PA Linearization

Talk to Our Engineers

For DPD algorithm design and PA linearization, contact our team.

Get in Touch