15.01.2013 Views

U. Glaeser

U. Glaeser

U. Glaeser

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

eal-time. Although floating-point processors give very effective calculation performance, the fixed-point<br />

processors are also often used, because they are characterized by a much lower supply energy consumption<br />

(e.g., 0.05 mW/MIPS for the TMS320C55x processors) and are, moreover, much cheaper [50].<br />

In order to facilitate communication with the host processor (computer), modern DSPs are equipped<br />

with a special interface—the host port interface (HPI) [49]. Using the HPI, the host processor has an<br />

access to the memory of the digital signal processor. Physical connection of processors uses 8- or 16-bit<br />

parallel data bus and several control lines. The HPI performs also a boot operation of the DSP. Due to<br />

this possibility an additional boot memory is not necessary. Selected DSPs provide also a glueless interface<br />

to the peripheral component interconnect (PCI) bus.<br />

The next interesting feature of modern DSPs is the use of a buffered serial port (BSP) [49]. BSP makes<br />

a high-speed communication with external devices, e.g., ADCs and DACs, possible. In relation to a typical<br />

serial port, BSP offers enhanced features, which allow for a direct read/write operations from/to the<br />

memory connected to the signal processor without any participation of its central processing unit (CPU).<br />

The BSP interrupts to CPU are generated after filling halves of the buffer. This machanism makes it<br />

possible to effectively cooperate with multichannel ADCs and DACs or to accumulate samples for FFT<br />

analysis. The described features, such as the HPI and the BSP, can substantially simplify the audio digital<br />

signal processing system (cf., Fig. 27.18) [25,28].<br />

Floating-point DSPs, with the support for the IEEE-754 standard of 32-bit floating-point format, are<br />

based on new architecture concepts in order to guarantee a very high computational efficiency. Among<br />

these concepts are:<br />

• Texas Instruments’ VelociTI, which is an advanced very long instruction word (VLIW) architecture<br />

[51],<br />

• Analog Devices super Harvard architecture (SHARC) with a single instruction multiple data<br />

(SIMD) facility [1].<br />

Effective utilization of these highly parallel architectures needs an efficient C-compiler and an assembly<br />

optimizer.<br />

Comparison of selected fixed- and floating-point DSP processors is presented in Tables 27.2 and 27.3<br />

[1,33,50,51].<br />

© 2002 by CRC Press LLC<br />

TABLE 27.2 Main Features of Fixed-Point DSPs<br />

Processor Family<br />

Bus<br />

Interface (bits)<br />

Instruction<br />

Rate (MHz) Core MIPS<br />

RAM<br />

(bits)<br />

ROM<br />

(KWords)<br />

ADSP-218x 16 33−75 33−75 128 K−1.6 M —<br />

DSP56300 24 66−160 66−160 120 K−1.2 M 0−80<br />

DSP56600 16 60−104 60−104 384 K−1.2 M 2−128<br />

TMS320C54x 16 40−160 30−532 80 K−3.1 M 0−48<br />

TMS320C55x 16 160−200 320−400 2.5 M 16<br />

TMS320C62x 32 150−300 1200−2400 1−7 M —<br />

TMS320C64x 32 400−600 3200−4800 8.5 M —<br />

Host<br />

computer<br />

DSP processor<br />

CPU<br />

HPI Internal BSP<br />

memory<br />

FIGURE 27.18 A digital audio signal processing system.<br />

multichannel<br />

audio codec

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!