GISdevelopment.net ---> AARS ---> ACRS 1996 ---> Digital Image Processing

Advanced Digital Signal Processing Techniques For Image Processing In Remote Sensing

Amal G. Punchihewa
Seniro Lecturer
Department of Electronics & Telecommunication Engineering
University of Moratuwa
Director
Sri Lanka Television Training Institute


Abstract
This paper describes the evolution of the microprocessor architecture from Von Neumann architecture to advance processor architectures. Intensive signal processing for remotely sensed data can be performed quite speedily with special microprocessor architectures such as Harvard architecture. Until the Pentium processor come up with the features of Harvard architecture, all general purpose processors were based on Von Neumann architecture. Certain bottle necks in the Von Neumann architecture are overcome through the Pentium having two cache memories; one for codes and other the for data.

Introduction
In remote sensing, the data captured by the satellite have to be processed before they transform into a useable from. These were mainly done by software programmes that were written to process the data in a computer. Due to the high quantum of data and the extensive processing required, it takes quite a long time to output data in a useable from. Since data mainly contain on images, signal processing that is done is called image processing. This paper tries to explain the potential of computers that are based on general purpose processors which have high processing power.

Since the invention of Von Neumann architecture, we have been using this architecture for general purpose microprocessors for nearly three decades. During the last two decades special architectures were developed for very specific applications such as digital signal processing in the audio and video industry. However these were restricted only to those applications and not incorporated into general processors until the Pentium adopted certain of their features. These processors can be used to process image data from satellite.

1. Image Processing
To develop a map using remote sensed row data have to be processed using algorithms. These algorithms are software programmes which may be a filter, an edge detection or image enhancement. Due to large quantum of data per iamge will take a long time to process an image.

In image processing simple operations such as addition subtraction division by 2 etc. are carried out repeatedly. Therefore in signal processing smaller set of instructions are repeatedly used. However in a general purpose processor we find only one processing element in the microprocessor. This is a bottle neck in signal processing with processors of fourth generation and below.

2. Conventional Microprocessor Architecture
Conventional microprocessors use the Von Neumann architecture. This is the first and most popular architecture of microprocessors in use for the past three decades. It consists of three phases or cycles where "stored programmes" and data are retrieved, interpreted and executed. This architecture has so many bottle necks and most fast enough for I today's applications.

2.1 Phases or three cycles of the Von Neumann Architecture
A microprocessor based on the Von Neumann architecture has to go through three stags viz.
  1. Fetch
  2. Decode
  3. Execution
In the etch stage an instruction or data is taken into a register from the memory, which may be outside or inside the CPU. If is an instruction, then it is decoded to perform the required function Then the function is performed in the execution process.

2.2 Shortcomings in the Conventional & Von Neumann Architecture
The basic Von Neumann architecture has one serious disadvantage: everything happens consecutively. Before any one operations completed, many steps (often very many) have to be completed. There are many shortcomings or bottle necks in the Conventional Architecture. They are,
  1. Only one process (fetch, decode or execution) takes place at a clock cycle.
  2. Only one processing element (one ALU)
  3. Limited data paths
These will limit the processing power of the processor. The performance of a micro-processor depends on word length (number of bits in a data) clock speed (frequency), internal bus architecture and the interfaces to peripherals. Maximum clock speed depends on the fabrication technology. Execution time is used to assess the performance of a microprocessor. Microprocessor has an execution time,

Execution time = (Number of instructions x (Average number of cycles/instruction)x (cycle time)

3. Digital Signal Processing
Microprocessors were mainly used for number crunching in financial departments. Today they are used for many control and decision-making processes. In the future they well be enquired to process numbers while executing algorithms for processing signals of audio and video. To process audio and video signals in real time, special processor architectures are required with special features. Some o these special architectures are listed in the next chapter. Today, most computers are equipped with Video Signal Processors (VSP) and Audio Signal Processors (ASP) of processing video and audio efficiently.

It is important to understand how do digital signal processors differ from micro-processors? In the first place a signal processor has to be capable of performing very large number of operations per unit time. The exact number depends directly on the bandwidth of the signal to be processed. At present the most advanced signal process can handle signals with the bandwidth o f up to several tens of MHz. One of the most common digital operations is the multiplication of pairs of numbers. From two sequences and addition of the product operation of this type are found in algorithms for filtering, correlation, spectral analysis etc. to perform these operations at a reasonably high speed the arithmetic unit must include a multiplier/accumulator combination.

4. The Architecture of a Digital Signal Processor
The basic Von Neumann architecture of the Conventional architecture has many limitations in signal processing .New technologies such as real-time video processing in multimedia are demanding high processing power from the microprocessors. They need to have many processing elements. These multiple processing elements will enable parallel processing. Therefore average instructions/per second will operate increase. This is usually measured in MIPS (million instructions per second ) and MOPS (million operations per second)

One way to overcome executing time barrier is to execute more than one instruction at a time (per lock cycle). Super scalar microprocessor have more than one processing element to process the data. General architecture for high performance microprocessor for signal processing are having super scalar, super pipe lined architecture. To achieve parallel processing (intra-processor parallelism), number of processing elements within I a microprocessor should be increased. In conventional general purpose microprocessor only one arithmetic and logic unit an be found. However in the Pentium processor, there are two ALUSs, enabling the processor to carry out pipe line processing. That is, two ALUs could perform two operations during a single machine (clock) cycle. This is a anther feature that is found in special microprocessor architectures which can be used for digital processing in audio and video application.

5. Harvard & Country Architectures
In Harvard Architectures data and instructions (code) are stored in two separate memories. This will enable to fetch data and code simultaneously. There are two separate connection for data and control information. This will enable to fetch data code simultaneously.

In country architectures, many processing elements are available with wider data paths. Counties are the separate processing elements with certain memory. In general, digital signal processing

For images and audio, simple operations such as addition, subtraction & average (division by 2) to be performed repeatedly for a data steam. Therefore in country architectures each processing element will perform a specific task and to due to many processing element being present in the processor, it is possible to perform many operations per clock cycle or per second.

6. Conclusions
There will be many advanced processor architectures that are used in general purpose computer for image processing. Since such advanced architectures are used in general purpose processors they will be affordable for the general computer users. This will enable remote sensed data to be processed on general purpose computers.