The DSP processor landscape is changing in many ways. For example, in years past, vendors offered numerous "general-purpose" DSPs intended to serve a wide range of applications. Today, many DSP families are focused on certain types of digital signal processing applications, such as control loops or audio equipment. In this article, we'll take a look at the current mainstream choices in DSP processors, and describe their key target applications and competitors.
ADI, Freescale, and TI
There are three main vendors of DSP processor chips: Analog Devices, Freescale, and Texas Instruments. In this article we'll cover their most commonly used processor families. We've grouped the processors into three categories: Low-cost fixed-point, high-performance fixed-point, and floating-point. These are very rough classifications and a few of the processors could reasonably be put into two categories; we'll note this where relevant.
Low-Cost Fixed-Point:
Freescale and Texas Instruments both offer multiple low-cost fixed-point processors. (Analog Devices also offers a low-cost fixed-point chip family, the ADSP-21xx, but this family is no longer being actively developed. It has largely been supplanted by ADI's Blackfin family, which is discussed in the section on high-performance fixed-point chips.)
In general, low-cost fixed-point DSPs are not particularly fast; they operate at modest clock speeds (typically 350 MHz or less—often much less) and they're mostly single-MAC devices that closely resemble the traditional DSP architectures of the early 1990's. (The multiply-accumulate – or MAC -- is a key operation in many signal processing algorithms. While MAC throughput isn't always a good predictor of signal processing performance, it is still a feature of interest when comparing DSP processor capabilities.)
Today, many modern embedded CPUs are actually faster than low-cost fixed-point DSPs. But in signal processing applications, embedded CPUs typically can't compete with DSP processors on power and cost efficiency, and they usually lack the specialized on-chip integration and development tools needed for signal processing and motor control applications. There are many embedded applications where speed is not the most important metric (such as motor control). For these applications, low-cost fixed-point DSPs are often a good choice because of their power and cost efficiency and their specialized integration and tools.
Freescale DSP563xx
Freescale's DSP563xx is currently the only mainstream 24-bit fixed-point processor available. The '563xx primarily targets high-fidelity audio applications, where its wider data word can yield better audio fidelity (relative to 16-bit fixed-point chips). The '563xx is a close descendent of the 24-bit DSP560xx, which was launched in the late 1980's as Freescale's (then Motorola's) first DSP processor. The 24-bit data word width was unusual then, as it is now. It helped win numerous audio equipment designs for the 5600x and 563xx families, and those early wins are a key reason why the family is focused on audio equipment today.
The fastest DSP563xx family members run at 275 MHz at 1.6 Volts, and pricing ranges from about $4 to $45. (All prices in this article are for 10K quantities.) Unlike most fixed-point DSPs, the '563xx often competes with 32-bit floating-point processors, which also offer good audio fidelity and have the advantage of an easier software development model. The trade-off is that floating-point chips often cost more and use more power.
Freescale DSP5685x
Freescale's DSP5685x 16-bit fixed-point processor family targets automotive, motor control, and digital power supplies, among other applications. This processor falls into the category of "digital signal controllers," meaning that it combines attributes of a traditional DSP processor with those of a microcontroller, and has peripherals suitable for control-loop applications, such as PWM. (The term "digital signal controller" is fairly new but has been adopted by a number of vendors.) Members of the DSP5685x family operate at up to 120 MHz, with pricing from $3 to $20.
Freescale also offers a related chip family, the MC56F83xx, that is based on the same core as the DSP5685x but contains flash memory and operates at a maximum clock speed of 60 MHz.
Texas Instruments TMS320C28x
The Texas Instruments TMS320C28x is a family of 32-bit fixed-point processors. Like Freescale's DSP5685x, the 'C28x is marketed as a digital signal controller, and includes family members with flash memory and control-oriented peripherals (such as CAN and PWM). The 'C28x's 32-bit fixed-point data word is unusual in DSP processors. It's a key differentiator because control-loop applications can be sensitive to dynamic range and precision, but are typically too cost-sensitive to use 32-bit floating-point chips. The 'C28x targets motor control and digital power applications.
The 'C28x executes at up to 150 MHz; it's one of the faster DSCs available. Chip pricing ranges from $3 to $14. TI also offers a lower-cost, lower-performance control-oriented architecture, the TMS320C24x, which uses a 16-bit fixed-point architecture and runs at up to 40 MHz.
Texas Instruments TMS320C55x
Texas Instruments' TMS320C55x chip family was introduced in 2000 as a successor to the widely used TMS320C54x. The 'C55x has two MAC units (compared to one on the 'C54x and other processors in the low-cost group) and can execute up to two instructions in parallel. It also has a much higher top clock rate than the 'C54x: 300 MHz rather than 160. The TMS320C55x is marketed for a wide range of applications, particularly those that require a combination of low cost, moderate DSP performance, and high energy efficiency, such as portable audio players and other consumer applications.
Like the other chip families in this group, the 'C55x is essentially a traditional, general-purpose DSP, but it is the only one in the group that can execute two instructions per cycle. It offers moderate DSP performance at low power and low cost, with chip pricing ranging from $4 to $17.
TI also offers an upgraded version of the 'C55x core, called the 'C55x+, that runs at a higher clock speed (up to 500 MHz) and has an enhanced microarchitecture. The C55x+ is available only in application-specific chips for wireless handsets—it's not intended to serve a wide range of applications. Based on its BDTImark2000™ score, the 'C55x+ offers significantly faster DSP performance than the 'C55x.
|
