Towards Low-Latency FFT Hardware Architectures
Timing of the Low-Latency 64-Parallel 4096-Point Memory-Based FFT

Towards Low-Latency FFT Hardware Architectures

One of the key topics in which we are working nowadays is the design of low-latency FFTs. This is an area that has barely been developed in the literature but, now, with the low-latency needs of digital communication, it becomes a key topic in our field.

The first work that we have proposed in this area is explained in the following journal paper, which appears in the IEEE Transactions on Circuits and Systems I this month of October:

Z. Kaya and M. Garrido, "Low-Latency 64-Parallel 4096-Point Memory-Based FFT for 6G," in IEEE Trans. Circuits Syst. I, Reg. Papers, vol. 70, no. 10, pp. 4004-4014, Oct. 2023.

The paper is an excellent work by Zeynep Kaya where she expands the parallelization in memory-based architectures, being the first time that a parallelization of 64 branches is achieved in this type of architectures. This high parallelization leads to a fast computation of the FFT that results in a 4096-point FFT architecture that carries out the calculations on an FPGA in only 2.7 us.

As usual, the paper is available as Open Access from IEEE Xplore. Additionally, we provide the code of the implementation in our GitLab repository FFTcores.

Our next steps towards low-latency FFTs will be to reduce the latency in other types of FFT architectures, as well as to provide new architectures specifically designed for low latency.

For companies that are interested in low-latency FFT architectures, it is the right time to contact us, as we expect to protect some of our desings with patents.

Our work on low-latency FFTs is carried out in the frame of the project "RAFFTING: Realizing Advanced FFT Implementations for 6G", PID2021-126991NA-I00, funded by the Spanish Ministry of Science and Innovation, the AEI, and the European Regional Development Fund through the call "Proyectos de Generación de Conocimiento, 2021".

要查看或添加评论,请登录

Mario Garrido的更多文章

  • Playing with the Data Order in FFTs

    Playing with the Data Order in FFTs

    An interesting topic when designing fast Fourier transform (FFT) architectures is the data order. When we want to…

  • Good News from DCIS 2024!

    Good News from DCIS 2024!

    Yesterday we were informed that the three papers that we submitted to the conference DCIS 2024 have been accepted to be…

    1 条评论
  • Further Improvements in MSC FFTs

    Further Improvements in MSC FFTs

    At the beginning of this year I wrote a post on the multi-path serial commutator (MSC) FFT, explaining the first…

  • Even More Efficient FFT Cores

    Even More Efficient FFT Cores

    We continue developing highly optimized FFT cores that beat the state-of-the-art again and again. With this aim, we…

    1 条评论
  • What is the limit in pleasing reviewers to get our papers accepted?

    What is the limit in pleasing reviewers to get our papers accepted?

    This week we have had a very interesting anecdote that has made me reflect on the paper review process. For those who…

    5 条评论
  • Towards Solving the Non-Power-of-Two FFT Challenge

    Towards Solving the Non-Power-of-Two FFT Challenge

    A few months ago, I wrote a post about the challenge of designing non-power-of-two (NP2) fast Fourier transform (FFT)…

  • The MSC FFT

    The MSC FFT

    Today I want to introduce the multi-path serial commutator (MSC) FFT. This type of architecture is the parallel version…

  • Is it Possible to Design Efficient Non-Power-of-Two FFT Hardware Architectures?

    Is it Possible to Design Efficient Non-Power-of-Two FFT Hardware Architectures?

    The challenge of designing non-power-of-two (NP2) FFT architectures has existed for decades. However, nowadays there is…

  • FFT Cores

    FFT Cores

    We have started a new Open Science initiative to make our research on fast Fourier transform (FFT) algorithms and…

    2 条评论
  • Sometimes, the best solutions are, indeed, simple

    Sometimes, the best solutions are, indeed, simple

    My PhD student Pedro Paz Mongil came one day to my office asking me about the calculation of the arcsine and arccosine…

    3 条评论

社区洞察

其他会员也浏览了