Digital Signal Processor IC Options for Embedded Applications

Digital signal processor IC on an embedded baseboard.

Your telephone conversations, streaming videos, streaming music, smartphone camera, and much more wouldn’t operate as designed without a digital signal processor IC. In many newer systems, the algorithms that run important digital signal processing tasks are being integrated into SoCs, being run as software on an embedded OS instance, or are simply offloaded to the cloud (e.g., in cloud-connected embedded systems). These tasks need to be defined at the hardware level, software level, or both, and component selection will dictate the computational time and accuracy of the results.

Digital Signal Processor ICs vs. FPGAs

Application areas like 5G and edge computing are moving away from FPGAs for processing power in favor of custom SoCs and specialized digital signal processor IC components. Growth in this market is already projected to grow at similar rates as the FPGA market thanks to ease of programming and greater specialization, and less need for parallelization. Because of these changes in the market landscape and device requirements, it helps to compare digital signal processor IC options with FPGAs as they can perform the same functions, but in different ways and with different performance metrics.

Both types of components can perform fixed point and floating point arithmetic operations, they have similar footprints, and similar cost per arithmetic operation in some cases. However, they have different feature sets, different programming learning curves, and completely different levels of specialization. FPGAs highly provide customizable programming, while DSPs are intended for specialized signal processing applications (thus the name). Only some specialized mixed-signal FPGAs include ADC/DAC blocks, while most high-performance DSPs will include DAC/ADC blocks for interfacing with sensors and other instruments.

To summarize, when you need a processor that provides highly customizable with shareable resources, faster processing speed, and significant parallelization, you’ll have faster calculations and lower cost per MAC, the better choice is an FPGA. However, if speed is not the critical factor, and you need specific integrated features, you’re better off using a digital signal processor IC. At lower clock/MAC rates, you’ll see faster calculations with similar cost per MAC as an FPGA.

A digital signal processor IC will need to interface with other components via standard protocols.

Digital Signal Processor IC Benchmarks

There are some important guidelines to consider when selecting a digital signal processor IC:

• Multiply-accumulate (MAC) rate: This is related to the following point, but it also depends on the bit depth for your fixed or floating point operations. This is the primary benchmark used to assess the performance of a digital signal processor IC.
• Clock/Instruction rate: Instruction rates are normally measured in MIPS (mega-instructions per second) or as a cycle time value.
• Performance against benchmark algorithms: Specific measurements of calculation times for benchmark algorithms gives you a benchmark with which to compare your own algorithms. Two common benchmark algorithms are finite impulse response filtering and FFT.
• Access to integrated peripherals: These ICs typically offer multiple programmable I/Os and interfaces on a single package for interfacing with other components. Pay attention to the signalling protocols available in your IC.
• Cost: one of the advantages of these components is their higher performance for signal processing tasks with lower cost than comparable MCUs or FPGAs (see above). There is often little advantage to overbuying unless you need access to more external peripherals or interfaces.

High Performance Digital Signal Processor IC Options

These options provide faster, more accurate computations compared to their FPGA counterparts at similar clock rates and with similar cost. The programming learning curve is also easier for these components, which helps many designers get new products into production faster than when an FPGA is used.

Texas Instruments, TMS320C6720BRFP200

The TMS320C6720BRFP200 from Texas Instruments is a low-cost digital signal processor IC that supports 32-bit fixed point, 32-bit floating point (single precision), or 64-bit floating point (double precision) computations. Some ideal applications for this component include high performance audio systems (e.g., real-time effects, audio synthesis, instrument modeling, encoding/broadcast), medical imaging (e.g., 3D tomography and image processing), biometrics, and other applications requiring specialized signal processing tasks.

This component does not include ADC/DAC blocks, although it does include 2 SPI and 2 I2C interfaces for connecting to external ADC/DAC components. This component also includes a Universal Host-Port Interface (UHPI), where an external host CPU can access memories on the component in parallel. Unlike some other digital signal processor ICs, there is some level of parallelization in computations:

At 350 MHz, the CPU is capable of a maximum performance of 2800 MIPS/2100 MFLOPS by executing up to eight instructions (six of which are floating-point instructions) in parallel each cycle. The CPU natively supports 32-bit fixed-point, 32-bit single-precision floating-point, and 64-bit double-precision floating-point arithmetic [from the TMS320C6720BRFP200 datasheet]

Functional block diagram for the TMS320C6720BRFP200 digital signal processor IC. From the TMS320C6720BRFP200 datasheet.

Analog Devices, ADSP-21161NCCAZ100

The ADSP-21161NCCAZ100 from Analog Devices is another digital signal processor IC that targets audio, video, medical, and industrial applications. This BGA component provides many more integrated peripherals, including 1 Mbit integrated SRAM, 16 Tx/Rx streams via I2S, an SPI bus, and a JTAG interface. It supports 32-bit fixed point, 32-bit floating point (single precision), and 40-bit floating point (extended precision) data formats at up to 660 MFLOPS.

As a benchmark evaluation algorithm, this component completes a 1024 point complex FFT calculation in just 92 μs and a finite impulse response filter at 5 ns per tap (100 MHz instruction rate). This makes the processor ideal for real-time audio and image processing applications. Calculation times for important benchmark algorithms are shown below.

ADSP-21161NCCAZ100 performance against benchmark algorithms. From the ADSP-21161NCCAZ100 datasheet.

Texas Instruments, 66AK2E05XABDA4

The 66AK2E05XABDA4 digital signal processor IC carries higher cost than the TI component shown above, but it provides much faster processing speed and access to many more peripherals. It can also interface with 2 USB 3.0 interfaces and 2 PCIe peripherals. It also provides 32x GPIO, 2x UART, and 3x SPI interfaces, as well as 1 GBE and 10 GBE Ethernet. Everything is built on a quad-core ARM A15 with 1.4 GHz clock rate. For processing capabilities, this component provides 32-bit fixed point (38.4 GMACS/Core @ 1.2 GHz) and floating point (19.2 GFlops/Core @ 1.2 GHz) calculations. The 66AK2E05XABDA4 includes one 64-bit wide, 1.5-V DDR3 SDRAM EMIF interface.

Functional block diagram for the 66AK2E05XABDA4 digital signal processor IC. From the 66AK2E05XABDA4 datasheet.

When you’re looking for a digital signal processor IC, you’ll find all the parts shown above and plenty of other options on Octopart. You’ll also find many options for supporting components for your next PCB.

Stay up-to-date with our latest articles by signing up for our newsletter.