Optimal Radix-2 FFT Compatible Filters for GFDM

Ahmad Nimr, Maximilian Matthé, Dan Zhang, Gerhard P. FettweisIEEE Communications Letters 2017 · OpenReview

TL;DR

This paper introduces a novel filter design for Generalized Frequency Division Multiplexing (GFDM) that allows for efficient, radix-2 FFT-based implementation even with an even number of subcarriers and subsymbols. By analytically deriving the optimal filter shift, they minimize the condition number of the GFDM modulation matrix, leading to improved performance and reduced implementation complexity.

What Problem Does This Paper Address?

Generalized Frequency Division Multiplexing (GFDM) is a promising multi-carrier waveform for future wireless systems due to its flexibility. However, a key challenge in GFDM is the design of the pulse-shaping filter. Previous research showed that using real, symmetric filters with an even number of subcarriers (K) and subsymbols (M) leads to a singular or near-singular modulation matrix. This ambiguity limits the number of data symbols that can be reliably transmitted and, crucially, prevents the efficient use of radix-2 Fast Fourier Transform (FFT) algorithms, which are fundamental for low-complexity implementations in conventional multicarrier systems like OFDM. This paper addresses this limitation by proposing a filter design that circumvents these issues, enabling efficient GFDM processing for a wider range of parameters.

Methods

The core of the paper lies in an analytical approach based on the Discrete Zak Transform (DZT) to characterize the GFDM modulation matrix. The authors decompose the modulation matrix and leverage the DZT to derive a closed-form expression for its condition number. They then introduce a fractional shift in the sampling of the continuous frequency response of the pulse-shaping filter. This "sampling offset" (denoted by $\lambda$) is the key innovation. By carefully choosing this offset, they can avoid the problematic zeros in the DZT of the filter, thus ensuring a non-singular modulation matrix even for even M and K. The paper analytically derives the optimal $\lambda$ that minimizes the condition number. Furthermore, they analyze two specific filter families: ISI-free filters without and with matched filtering, and derive their respective condition numbers. Finally, they numerically evaluate the performance of their proposed design using the Signal-to-Interference Ratio (SIR) and Noise Enhancement Factor (NEF) for Zero-Forcing (ZF) and Matched Filter (MF) receivers.

Datasets

Not discussed in the paper. The paper focuses on theoretical analysis and numerical evaluation using simulated filter responses.

Key Insights and Ideas

  • Fractional Sampling Offset is Key: The most significant insight is that a fractional shift in the filter's frequency response sampling ($\lambda$) can fundamentally alter the properties of the modulation matrix, specifically its condition number. This provides a powerful tool to overcome limitations imposed by filter symmetry and even parameter counts.
  • Radix-2 FFT Compatibility Achieved: By ensuring a non-singular modulation matrix, the proposed filter design directly enables the use of efficient radix-2 FFT algorithms for GFDM processing, a major step towards practical and low-complexity GFDM implementations.
  • Analytical Condition Number Derivation: The ability to analytically derive the condition number of the modulation matrix is crucial. It allows for the precise identification of optimal filter parameters and provides a theoretical basis for understanding performance trade-offs.
  • Trade-off between Filter Design and Condition Number: The paper demonstrates that different filter design criteria (e.g., ISI-free vs. ISI-free with matched filtering) lead to different condition numbers, highlighting the interplay between filter properties and system performance.

Key Results

  • A closed-form expression for the condition number of the GFDM modulation matrix is derived using the DZT.
  • An optimal fractional sampling offset ($\lambda$) is analytically determined to minimize the condition number, enabling GFDM with even M and K.
  • The proposed filter design allows for the use of radix-2 FFT operations, significantly reducing implementation complexity.
  • Numerical evaluations show that the optimal $\lambda$ leads to improved NEF and SIR performance for both ZF and MF receivers. For instance, Figure 1 shows that the condition number is minimized at $\lambda=0$ for odd M and $\lambda=0.5$ for even M (when K is also even). Figure 3 indicates that an optimal $\lambda$ preserves the trend of NEF increasing with M while maintaining good SIR.

Limitations and Future Work

  • Specific Filter Families: While the paper analyzes ISI-free filters, the derived optimal $\lambda$ is generally applicable. However, the specific performance gains and the exact optimal $\lambda$ will depend on the chosen basis filter.
  • Real-valued Filters: The analysis primarily focuses on real-valued filters. The behavior with complex filters might differ.
  • AWGN Channel: The performance evaluation is conducted in an Additive White Gaussian Noise (AWGN) channel. Performance in more complex channel conditions (e.g., fading) would require further investigation.

The authors suggest that their work has a significant impact on implementation complexity and narrows the design space of GFDM as a flexible waveform generator. Future work could explore the application of this design in more realistic channel models and investigate its performance with different types of pulse-shaping filters.

Summary updated

Abstract

For a linear waveform, a finite condition number of the corresponding modulation matrix is necessary for the waveform to convey the message without ambiguity. Based on the Zak transform, this letter presents an analytical approach to compute the condition number of the modulation matrix for the multi-carrier waveform generalized frequency division multiplexing (GFDM). We further propose a filter design that yields non-singular modulation matrices for an even number of subcarriers and subsymbols, which is not achievable for any previous work. Such a new design has a significant impact on implementation complexity, as the radix-2 FFT operations for conventional multicarrier waveforms can readily be employed for GFDM. Also, we analytically derive the optimal filter that minimizes the condition number. We numerically evaluate the signal-to-interference ratio (SIR) and noise-enhancement factor (NEF) for matched filter (MF) and zero-forcing (ZF) GFDM receivers for the proposed design, respectively.
BibTeX
@article{nimr2017optimal,
  author = {Ahmad Nimr and Maximilian Matthé and Dan Zhang and Gerhard P. Fettweis},
  title = {{Optimal Radix-2 FFT Compatible Filters for GFDM}},
  journal = {IEEE Communications Letters},
  year = {2017},
  url = {https://openreview.net/forum?id=PYVetiSl6D4},
}
Figure