Unlocking the Potential of AMD Versal AI Engines
April 15, 2025
Sponsored Blog
As the complexity of signal processing applications continues to grow, engineers need to use all available heterogeneous processing elements to meet performance goals. AMD Versal™ Adaptive SoCs address this challenge by integrating programmable logic, hardened compute resources, a high-speed network-on-chip (NoC), and a scalable AI Engine (AIE) array—all within a unified architecture.
At Fidus Systems, we specialize in FPGA and ASIC design, verification, and validation for complex embedded applications. In one of our recent projects, we engineered a high-throughput implementation of the Multiple Signal Classification (MUSIC) algorithm within the Versal AI Engine (AIE) tile array. This work, completed in collaboration with AMD, demonstrates Fidus’s ability to translate algorithmic complexity into efficient, hardware-accelerated solutions, driving performance, scalability, and integration in real-world embedded systems.
What Is MUSIC and Why Does It Matter?
MUSIC is a subspace-based signal processing technique widely used for Direction of Arrival (DOA) estimation, modal analysis, and spectral estimation. Its ability to distinguish multiple signal sources with high resolution makes it well-suited for radar, sonar, wireless communication, and structural health monitoring applications.
But MUSIC is also computationally intensive. It requires QR decomposition, Singular Value Decomposition (SVD), and spectral scan evaluation across finely spaced angles. These operations are a natural fit for parallel acceleration but must be precisely orchestrated to meet embedded system constraints.
Mapping MUSIC to Versal AI Engines
Unlike a proof-of-concept or demo, this design was developed as a real implementation pipeline—one that could scale and integrate into a broader system. The Fidus engineering team applied the same rigor we bring to every project: from system modeling and kernel optimization to integration, validation, and delivery.
We applied the following architecture decisions:
- Loop unrolling and deep pipelining across tiles to break down QRD, SVD, and spectrum search
- Memory-local data sharing to minimize bandwidth and latency between stages
- C/C++ kernel-level optimization to eliminate redundant operations and improve tile utilization
Using the AMD VCK190 evaluation board, our engineers deployed the solution and validated it with a MATLAB-based hardware-in-the-loop (HIL) setup. Snapshot data was streamed into the design, processed on-chip, and the DOA output visualized in real time.
Why Versal Is the Right Platform for MUSIC
The Versal AI Engine array offers architectural features that directly align with the needs of compute-bound DSP algorithms like MUSIC:
✔ Parallel VLIW + SIMD execution across a scalable tile array
✔ Deterministic data movement and local memory sharing
✔ Tight integration with PL, PS, and high-speed I/O via NoC and DMA
✔ Seamless software/hardware co-design with C/C++ kernel support
These capabilities allowed us to offload MUSIC from traditional processors or PL-based implementations while achieving significant performance gains.
At Fidus, we specialize in high-performance FPGA designs, AI Engine, and embedded system design. If you’re tackling a complex project, book a consultation to see how we can help.
👉 Watch the full Webinar: Offload Multiple Signal Classification (MUSIC) to AMD Versal™ AI Engines
*Special thanks to Bachir Berkane, System and Algorithm Architect, and Peifang Zhou, Senior Embedded Software Designer at Fidus Systems, along with Mark Rollins, Ph.D., Senior Manager, Technical Marketing at AMD, for their contributions to this webinar. Scott Turnbull is the Chief Technology Officer (CTO) at Fidus Systems, driving technical innovation and guiding the company’s strategic direction in embedded system design. Fidus Systems is a leading provider of high-performance FPGA, AI, and embedded system design solutions. As AMD’s Partner of the Year, Fidus specializes in complex hardware acceleration, AI engine optimization, and high-speed data processing. With over 25 years of expertise, we help companies innovate faster, optimize performance, and bring cutting-edge technology to the market. Learn more at www.fidus.com or book a design consultation with our experts today.
