Engineering Design NIDisk Betriebsanweisung PDF herunterladen (Seite 32)

SIGNAL/RTS APPLICATION NOTES

8 Application Note 11

Issues in Data Storage and Exchange

This section provides background on data storage issues that affect data exchange between

software systems and hardware platforms. A basic understanding of these issues is important

to exchange data accurately and reliably.

Text vs. Binary format

Data files can be stored and exchanged in either text (ASCII) or binary format. Text format

is the most universal in that virtually all hardware platforms and software systems can read it,

however it imposes two limitations: 1) it requires considerably more storage (up to 10 times

more) for the same data and 2) numerical precision is limited to the data format chosen at the

time of writing (for example, F12.4 format limits precision to 4 decimal places). By contrast,

binary format is more space-efficient and numerical precision is determined by the data

storage format inside the computer, which is then maintained in the data file. For these

reasons, precision transfers of numerical data should usually be performed via binary files.

Integer vs. Floating Point format

Binary data in turn can be stored in either integer or floating point (real) format. Integer

format stores the (integer) data in direct binary form, e.g., 1001 for the number 9. It has

complete accuracy but a limited numerical range - e.g., a 16-bit integer can only represent

integers between -32768 and +32767. Floating point format refers to the coding of real

numbers, i.e., numbers which can assume non-integer values such as 1.234567. In this

format, a certain number of bits (the "mantissa") represent the numerical value without regard

for decimal point (e.g. 1234567 in the above), while the remaining bits (the "exponent")

indicate the position of the decimal point (i.e. 10 ** -6 in the above example). The number of

mantissa bits limits the precision of the format, while the number of exponent bits limits its

dynamic range (largest positive and negative number). IBM PC's use a standard 32-bit

floating point representation (IEEE 754) which has 1 sign bit, a 24-bit mantissa (1 bit is

implicit), and an 8-bit exponent. This provides a mantissa resolution of 2 ** 24 = 10 ** 7, or

7 decimal places, and an exponent range of 2 ** 8 = 128 (i.e., the binary "decimal point" can

be moved up to 128 binary places in either direction). Converting to powers of ten, the

dynamic range is then 2 **

±128 = 10 ** ±37.

The two numerical formats serve quite different purposes. Raw acquisition data is normally

maintained in the original integer format delivered by the digitizer, to avoid introducing even

minor "noise" by converting the integers to the less precise floating point format. Conversely,

the results of calculations are almost universally stored in floating point format, since their

precision and range vary widely.

1 2 ... 27 28 29 30 31 32 33 34

Keine Kommentare

Engineering Design NIDisk Betriebsanweisung Seite 32