Crunching the Same Numbers on Different Architectures
DSP, microcontroller or a general purpose CPU? What are the major factors affecting design choices in embedded systems today, and how do they compare in terms of computational speed?
Since 1978 DSP chips were invented to cope with some limitations of existing CPUs. Audio processing has always been challenging because of inherently high demand for both throughput and low latency processing features. This led the chip manufacturers to develop specific architectures for such challenging tasks.
During the years the increasing demand of computation led also to develop on-chip hardware accelerators for specific tasks like FIR, IIR and DMA, making the machines more and more performant generation after generation. Hardware acceleration is great, but has a drawback: the design then becomes more and more architecture-locked, preventing a faster renewal of the OEM platforms and the embracement of new emerging technologies.
While the DSP world was evolving, also the general purpose CPUs made very big steps forward in terms of computational speed. They typically run with some high end operating systems like Linux that offer a wide set of ready to use features such as networking, USB, BT… and a huge number of available libraries you can use to fasten the audio application development cycle. This makes them very attractive in terms of flexibility.
In the middle, between specialised DSPs and general purpose CPUs there are microcontrollers. They integrate some other hardware that a product usually requires and provide low power operation. We will explore why they are one of the choices available and how they position in terms of computational performance.
The aim of this talk is to give an overview of the different architectures used in embedded audio signal processing, their advantages and disadvantages, and try to compare them with a holistic view. This will take into account computation power, ease of use, BOM cost, power requirements, expandability and give an overview of why all the architectures are still there in 2024 competing after so many years.
We will make use of real code for the comparison that will also highlight how the different chip architectures have an impact on code design.
Will there be any winner?
Marco Del Fiasco
Embedded Systems Architect
Del Fiasco Marco
I am a freelance embedded engineer helping audio companies to develop modern digital platforms and algorithms.
I have 15 years experience with many projects across the music industry, pro audio and car audio.
My expertise spans from DSP chips and microcontrollers to modern CPUs running hard realtime audio processing.
I developed several algorithms, peripheral drivers, libraries and frameworks making use of bare-metal, embedded RTOS and embedded Linux.