DSPs and modern device design

5 mins read

With almost any technology, the path to progress is chaotic. No matter how mundane the gadget in your hand might seem, it’s part of a market that is constantly iterating on itself.

Credit: Distinctive Images - adobe.stock.com

Designers are always reimagining, refining, and reinventing their products and, as a result, the devices we’re developing are having to accommodate an increasingly wide spectrum of inputs and signals. As new ideas take hold and our electronics become more capable of communicating both with us and with one another, we need them to be able to recognise and understand one another without missing a beat.

This is especially important in the era of the Internet of Things (IoT). If we’re developing products capable of intelligently processing data and performing tasks autonomously, they need to be able to recognise stimulus perfectly and respond to it appropriately.

Amidst the chaos of development, we also need to be able to rely on a common frame of reference for understanding those signals – technologies in common that help us to perform the basics of communication.

The 101 on DSP

One of these is digital signal processing, or DSP. DSP silicon has been very popular and pervasive in all our devices; we use it on a day-to-day basis without even realising it.

Just as it sounds, DSP is literally digitising the functions of analogue signal filters. You take real-world input and manipulate the maths behind that data in certain ways to achieve the desired output.

Analogue filters are well established – but so are their limitations. A physical filter is fixed once you build it, which means that components themselves can delay, and their values can drift. The longer you use one, the more the quality of its output gradually deteriorates.

DSP offers several advantages – the most obvious of which is that it’s not going to deteriorate in the same way. Encoding the filter digitally gives you a robustness that a physical alternative can’t necessarily match.

Furthermore, DSP can incorporate very specific effects that are more difficult through analogue equivalents. That’s not just ‘cleaning’ the audio to better isolate and understand the signal within, but also augmenting it: compression, modulation, equalisation, and so on. Once the process is complete, the device can output that signal however it’s intended.

The ubiquity of DSP, then, comes from its consistency and its efficiency: it’s a trusted means of being able to receive and process key information.

So, what does that consistency mean for modern device design? Why is DSP so important to the ways in which our electronics communicate with one another?

Well, the electronics market has massively diversified over the past few years. It’s not so long ago that semiconductor manufacturers could sink an immense amount of time and resource into chips for one specific market sector: personal computers, digital cameras, and so on.

These days, as more and more smart devices have been brought to the table and entirely new markets have sprung up, we’ve seen a diverse market of small niches emerge, each with unique requirements.

In an ideal world, every company designing a product has the money, time, and supply to cater to that niche with the perfect hardware. But with the sheer number of use cases and applications, that’s not realistic. Manufacturers have to work within their limits, finding the most bang for their buck on a set budget.

In that context, the ubiquity and utility of DSP is incredibly valuable. Engineers might not know exactly what component or budget is available to them at the start of a project – but they do know that DSP is a reliable, established means of building important functionality into their designs.

Given the breadth of technologies that need to be able to take advantage of DSP, however, it’s also important to consider exactly what sort of DSP solution you should be seeking and we’re talking about two general options here: a ‘pure’ DSP solution dedicated to the process and nothing else, and a general-purpose solution that can combine DSP with other functions.

While a pure DSP solution might function more efficiently or perhaps with more precision, the trade-off is that it would be horrendously inefficient at anything that isn’t DSP. Even integrating such a component, for example, is more complicated. That means a slower time to market, more dedicated hardware impacting your bill of materials, and a trickier technical process to bring everything together.

General hardware, by contrast, is likely to speak to that aforementioned balancing of cost and performance. It affords far greater integrability than a ‘dedicated’ DSP solution, with engineers able to combine other elements with DSP functionality. Some of the most effective solutions will enable engineers to code and recode entirely in software, further reducing the strain on components, cost, and time.

Above: xcore.ai provides low latency with highly deterministic performance, making it suitable for intelligent IoT applications (Credit: XMOS)

Artificial intelligence

Of course, this versatility is taken to the next level with artificial intelligence.

AI’s ability to augment the capabilities of DSP-enabled devices can transform the end product – something that a dedicated DSP would struggle with.

In simple terms, AI helps to inform the machine of the most sensible response to the signal that it’s receiving. It’s capable of enhancing a device’s ability to improve the quality and clarity of that signal. AI-enabled DSP can thus work with ‘dirtier’ signals than many conventional alternatives as they’re capable of ‘cleaning’ said signals more effectively.

A great example that impacts our daily lives is using voice to talk to devices. If you’re video conferencing, for example, it’s DSP technologies that take in our voices and apply signal processing algorithms. But with the advent of AI, they are usually coupled with machine learning algorithms that are sensitive to the environment you are in, and can remove external noises like wind, or a dog barking.

AI then enables DSP to communicate certain data, or insights, to other parts of the device, where it can be actioned appropriately. Some hardware eliminates the cloud almost entirely, with adequate intelligence enabling the on-device processing of data for improved privacy and performance.

So, an AI-enabled DSP offers an accessible, versatile, powerful means of capturing and processing data, and triggering a response to the insights within. But how are we going to see that manifest itself in 2024?

Interactivity in personal electronics is very likely to be the most widespread opportunity. Computers, TVs, wearables, and other devices where audio commands are increasingly common will perform at a much higher level, distinguishing words more effectively from background noise.

AI’s role in the audio ‘cleaning’ process makes that possible, while it can also enable more impressive functionality full stop. Speakers can be identified as different people tied to certain subscriptions or user accounts, for example, streamlining and accelerating the user experience.

We are likely to see similar advances in the automotive market, with a refinement of audio instruction taking place.

New voice interfaces and interactions will emerge, with DSP an important quality control element in the cockpit. Similarly, DSP can help to cancel out background noises, like high winds or road vibrations.

DSP is also an important component of the factory in which these vehicles are designed. In cleaning up the significant background noise in industrial settings, DSP can increase the range at which voice commands are recognised and actioned, making for a safer, more efficient factory floor. It can also improve the quality of data analytics on industrial machinery, identifying sounds or patterns that can indicate a need for pre-emptive maintenance.

These examples demonstrate how the principles of DSP, in combination with AI and IO, can be applied to great benefit across a myriad of industries.

In an ecosystem increasingly beset with interactive, AI-enabled electronics and machinery, the ability to communicate clearly with both humans and other devices is paramount.

We’ll see solutions evolve to be able to integrate all of these elements in a scalable, flexible way to reflect that need.

Ultimately, DSP is just an algorithm. But the capacity it has to improve our ability to interact with electronics in meaningful ways, given the number and variety of devices being developed, elevates it far beyond that.

Author details: Aneet Chopra, EVP Marketing & Product Management, XMOS