Category:
A
B
C
D
E
F
G
H
I
J
K
L
M
N
O
P
Q
R
S
T
U
V
W
X
Y
Z
0
1
2
3
4
5
6
7
8
9
Position:IcFull.com » IC Electronic information » Category: D

IC Electronic information

DSP VS Common MCU

In Electronic Infomation Category: D | on April 28,2011

Consider an instance of a digital signal processing, such as finite impulse response filter (FIR). For the use of mathematical language, FIR filters do a series of dot product. Take a number of inputs and EP2C5F256C8N datasheet and a vector sequence in the coefficients and EP2C5F256C8N price and input samples intercropping sliding window multiplication, and EP2C5F256C8N suppliers and then add up all the product to form an output sample.

Similar operation in digital signal processing large numbers of recurring, making this device is designed to provide dedicated support, contributed to a general purpose processor and DSP devices (GPP) of the split:

1 Dui multiplication intensive support

GPP are not designed for multiplication-intensive tasks, even some modern GPP, also require more instruction cycles to do a multiplication. The DSP processors use specialized hardware to achieve a single cycle of multiplication. DSP processors to handle the increased number of accumulator register the product and. Accumulator register is usually wider than other registers, increasing the extra bits as the result bits to avoid overflow. Meanwhile, in order to fully reflect the special multiplication - the benefits of cumulative hardware, almost all of the DSP instruction set contains an explicit MAC instructions.

2 memory architecture

Traditionally, GPP to use Feng. Neumann memory architecture. This structure, only one memory space by a group of bus (an address bus and a data bus) is connected to the processor core. Typically, once a multiplication occurs 4 times the memory access, spent at least four instruction cycles.

Most of the DSP uses Harvard architecture, the memory space is divided into two, respectively, stored procedures and data. They have two sets of bus connected to the processor core, allowing them access to the same time. This arrangement will double the bandwidth of the processor memory, more importantly, both the processor core to provide data and instructions. In this layout the next, DSP to achieve single-cycle MAC instruction.

There is a problem that GPP is a typical high-performance chip actually contains two caches, one for data, one command, they are directly connected to the processor core to speed up runtime access. Physically speaking, this dual-chip memory and bus structure is almost the same as a Harvard architecture. However, logically speaking, there are important differences between the two.

GPP using the control logic to determine which data and instruction words are stored in the on-chip high-speed cache, the programmer is not to be specified (and probably do not know.) In contrast, DSP using multiple chip memory and multiple-bus to ensure that every instruction cycle, memory access times. In the use of DSP, the programmer to explicitly control which data and instructions stored in the chip to memory. Programmers write programs, you must ensure that the processor can effectively use its dual bus.

In addition, DSP processors almost do not have the data cache. This is because the typical DSP data is the data stream. In other words, DSP processor for each data sample to do the calculation, you discarded, and almost no reuse.

3 zero-overhead loop

Knowing a DSP algorithm is a common feature that most of the processing time is * in the implementation of the cycle of small, it is easy to understand why most of the DSP has a dedicated hardware, for zero-overhead cycle. The so-called zero-overhead loop is the loop in the implementation of the processor, do not * have time to check the loop counter value, conditions shifted to the top of the loop, the loop counter by 1.

Contrast, GPPs recycling software. GPP with the transfer of certain high-performance hardware prediction, almost zero overhead hardware loop support to the same effect.

4 fixed-point calculations Calculated using fixed-point DSP

majority, instead of using floating-point. Although the application of DSP must pay great attention to the precise figures, with the floating-point should be much easier to do, but DSP, the cheap is also very important. Compared to the corresponding fixed-point floating-point machine to machine cheaper (and faster). But do not use floating-point machine to ensure the accuracy of figures, DSP processor instruction set and hardware support saturation calculations, rounding and shift.

EP2C5F256C8N datasheetEP2C5F256C8N suppliersEP2C5F256C8N Price

All right reserved:icfull.com © 2010-2016 Certificate