For always-on functions, such as voice control and activity classification, devices need to process inputs from sensors in real time and then perform operations based on the results. Digital signal processing (DSP) is the art of getting this done, but a digital signal processor is needed in order to do it near instantaneously and without overloading the device’s CPU.
What Is Digital Signal Processing?
Imagine you’re asking something from a smart speaker (like Alexa), such as a search of the nearest sushi restaurants. Analog-to-digital converters take your speech input and turn it into zeros and ones. These values are then transferred to a digital signal processor as an input, and the processor then analyzes these signals to turn them into words that can be processed. After processing, it will return a digital representation of its response to a waiting digital-to-analog converter that will output the information back to the user over a speaker.
(Source: CEVA)
Hypothetically, this same sequence of events could be done with analog signal processing, but it would be significantly more complex, and changes to the algorithm would require a lot of tweaking. Additionally, changes to the processing algorithm would require physical components to be optimized. By comparison, DSP code can be changed instantly without the large component upkeep. To add to its benefits, since DSP inputs are digital, they will give the same output for any given input every time. Physical analog components have anomalies that can change with varying operating conditions.
Where Is DSP Used?
DSP is used in many electronics applications. For example, a computer may use DSP to monitor security, transmit phone calls, compress videos or play a movie on a home theater system. In certain applications, the quality of the signal is enhanced to provide even more information and detail than what humans are able to sense. A computer that processes and enhances medical images is a good example of this.
DSP technology is often a key component of artificial intelligence (AI) applications, including natural language processing (NLP). NLP is used so that computers and devices can understand and analyze human language and behaviors by converting it into the “natural language”.
DSP technology is used across a variety of devices because the encoding and decoding techniques are standard. Unlike analog signals, which are prone to distortion, interference and even security breaches, DSP is a good option for high-speed applications requiring encryption, compression and rapid transmission.
Enabling Functions with Always-On DSPs
I’ve talked before about how a sensor hub can be used to offload the work of sensor fusion from the main CPU of your device, and a DSP offers the same benefit, allowing you to run multiple functions without overtaxing your primary CPU. For applications requiring always-on functions, embedded DSPs offer a cost-effective, low-power solution that can meet performance requirements.
Always-on functions operate in the background, and can be a combination of multiple sensor types, such as IMUs, voice or presence detection for more comprehensive context. Examples of always-on functions include:
- Pedometers and GPS in smartphones or wearables
- Lane assist or passenger seat detection in cars
- Voice control on smart speakers
In sensing applications, sensors are gathering information on light, sound, 3D movement and relative location. Some multi-axis sensors, including IMUs, perform their own data processing via sensor fusion to blend several inputs. A DSP can take this sensor fusion data and process it with additional sensor data (light, sound, etc) to tell a comprehensive story.
Most smart audio, video and imaging applications require at least some of these types of always-on voice control and object detection functions. For those functions to be executed, multiple sensor inputs need to be processed in real time. With a dedicated digital signal processor, these computations can be run in parallel with the CPU, so that many different functions can be carried out at the same time.
Dedicated processing is a common theme of digital signal processors and sensor hubs, which provide specialized functionality to elevate and accelerate the development of technology for the fast paced demands of tomorrow.
- Short overview of DSP:
- DSP uses digital signal processing to convert and analyze signals such as audio, video, voice, light, temperature, pressure or position, and then output usable data
- Analog converter takes this real-world information (such as light or sound waves) and turns it into a digital format (binary code); then, DSP technology processes this code and feeds the digitized information back out; this process is performed very quickly
- DSP is used in many electronic applications
- A computer may use DSP to monitor security, transmit telephone calls, compress video or play a movie on a home theater system
- In certain applications, the quality of the signal is enhanced to provide even more information and detail than what humans are able to sense – for example, a computer enhancing medical images
-
Analog signal processing is also possible, but the process is made much faster and more efficient with digital signal processing – improves speed and accuracy
- Analog signals were used traditionally for long distances, but are prone to distortion, interference and even security breaches
- Along with higher speed and accuracy, digital signal processing offers a number of benefits, including:
- More flexible hardware interpretation
- Easier to be used across a variety of devices because the encoding and decoding techniques are standard
- Encryption and compression help with security as well as efficient transmission and downloading
- By running a small always-on DSP, you can enable functions in the background:
- I’ve talked before about how a sensor hub can be used to offload the work of sensor fusion from the main CPU of your device, and a DSP offers the same benefit, allowing you to run multiple functions without overtaxing your primary CPU
- In sensing applications, sensors are gathering information on light, sound, or, if it’s a motion sensor, 3D movement and relative location; in AR/VR this might mean tracking hand motions; in robotics, this means mapping out surrounding objects and relaying that data to avoid crashes
- Some multi-axis sensors, like IMUs, do their own data processing via sensor fusion to blend several inputs, such as from a gyroscope and an accelerometer. A DSP can then process inputs (signals) from multiple sensors of different types, and this could include contextual motion data that’s already been processed via sensor fusion and is now being added to additional sensor data (such as light, sound, etc.) to tell a comprehensive story
- Always-on functions operate in the background, and can be a combination of multiple sensor types, such as IMU/Voice/other (e.g. presence detection) for more comprehensive context
- Example always-on functions: pedometers, GPS, lane assist or passenger detection in cars, voice control on TV remotes
- Most smart audio and video/imaging applications require at least some of these types of always-on voice control and object detection functions
- On a DSP, these computations can be run in parallel with the CPU, so that many different functions can be carried out at the same time
- Embedded DSPs can handle all of these always-on functions in real-time, which is critical to performance without slowing down the CPU or draining its battery – offers a more cost-effective, low-power solution.
Ceva