Intel® Integrated Performance Primitives (Intel® IPP) Developer Guide and Reference

ID 790148
Date 3/22/2024
Public

A newer version of this document is available. Customers should click here to go to the newest version.

Document Table of Contents

Glossary

adaptive filter

Colors specified by each pixel's coordinates in a color space. Intel® Integrated Performance Primitives for image processing use images with absolute colors.An adaptive filter varies its filter coefficients (taps) over time. Typically, the filter's coefficients are varied to make its output match a prototype “desired” signal as closely as possible. Non-adaptive filters do not vary their filter coefficients over time.

BQ

One of the modes, which indicates that the IIR initialization function initializes a cascade of biquads.

CCS

See complex conjugate-symmetric

companding functions

The functions that perform an operation of data compression by using a logarithmic encoder-decoder. Companding allows you to maintain the percentage error constant by logarithmically spacing the quantization levels.

complex conjugate-symmetric

A kind of symmetry that arises in the Fourier transform of real signals. A complex conjugate-symmetric signal has the property that x(-n) = x(n)*, where “*” denotes conjugation.

conjugate

The conjugate of a complex number a + bj is a - bj.

conjugate-symmetric

See complex conjugate-symmetric.

DCT

Acronym for the discrete cosine transform.

decimation

Filtering a signal followed by down-sampling. Filtering prevents aliasing distortion in the subsequent down-sampling. See down-sampling.

down-sampling

Down-sampling conceptually decreases a signal's sampling rate by removing samples from between neighboring samples of a signal. See decimation.

element-wise

An element-wise operation performs the same operation on each element of a vector, or uses the elements of the same position in multiple vectors as inputs to the operation. For example, the element-wise addition of the vectors {x0, x1, x2} and {y0, y1, y2} is performed as follows: {x0, x1, x2} + {y0, y1, y2} = {x0 + y0, x1 + y1, x2 + y2}.

FIR

Abbreviation for finite impulse response filter. Finite impulse response filters do not vary their filter coefficients (taps) over time.

FIR LMS

Abbreviation for least mean squares finite impulse response filter.

fixed-point data format

A format that assigns one bit for a sign and all other bits for fractional part. This format is used for optimized conversion operations with signed, purely fractional vectors. For example, S.31 format assumes a sign bit and 31 fractional bits; S15.16 assumes a sign bit, 15 integer bits, and 16 fractional bits.

IIR

Abbreviation for infinite impulse response filters.

in-place operation

A function that performs its operation in-place, takes its input from an array and returns its output to the same array. See not-in-place operation.

interpolation

Up-sampling a signal followed by filtering. The filtering gives the inserted samples a value close to the samples of their neighboring samples in the original signal. See up-sampling.

LMS

Abbreviation for least mean square, an algorithm frequently used as a measure of the difference between two signals. Also used as shorthand for an adaptive FIR filter employing the LMS algorithm for adaptation.

LTI

Abbreviation for linear time-invariant systems. In LTI systems, if an input consists of the sum of a number of signals, then the output is the sum of the system's responses to each signal considered separately.

MMX™ technology

An enhancement to Intel architecture aimed at better performance in multimedia and communications applications. The technology uses four additional data types, eight 64-bit MMX registers, and 57 additional instructions implementing the SIMD (single instruction, multiple data) technique.

MR

One of the modes, indicating the multi-rate variety of the function.

multi-rate

An operation or signal processing system involving signals with multiple sample rates. Decimation and interpolation are examples of multi-rate operations.

not-in-place operation

A function that performs its operation not-in-place takes its input from a source array and puts its output in a second, destination array.

polyphase

A computationally efficient method for multi-rate filtering. For example, interpolation or decimation.

CCS

A representation of a complex conjugate-symmetric sequence which is easier to use than the Pack or Perm formats.

Pack

A compact representation of a complex conjugate-symmetric sequence. The disadvantage of this format is that it is not the natural format used by the real FFT algorithms (“natural” in the sense that bit-reversed order is natural for radix-2 complex FFTs).

Perm

A format for storing the values for the FFT algorithm. The Perm format stores the values in the order in which the FFT algorithm uses them. That is, the real and imaginary parts of a given sample need not be adjacent.

saturation

Using saturation arithmetic, when a number exceeds the data-range limit for its data type, it saturates to the upper data-range limit. For example, a signed word greater than 7FFFh saturates to 7FFFh. When a number is less than the lower data-range limit, it saturates to the lower data-range. For example, a signed word less than 8000h saturates to 8000h.

sinusoid

See tone

Intel® Streaming SIMD Extensions

The major enhancement to Intel architecture instruction set. Incorporates a group of general-purpose floating-point instructions operating on packed data, additional packed integer instructions, together with cacheability control and state management instructions. These instructions significantly improve performance of applications using compute-intensive processing of floating-point and integer data.

tone

A sinusoid of a given frequency, phase, and magnitude. Tones are used as test signals and as building blocks for more complex signals.

up-sampling

Up-sampling conceptually increases the signal sampling rate by inserting zero-valued samples between neighboring samples of a signal.

window

A mathematical function by which a signal is multiplied to improve the characteristics of some subsequent analysis. Windows are commonly used in FFT-based spectral analysis.