Lucent Applications Library

DSP32C-AL-MSDOS
DSP32C-AL-SUN4
DSP3210-AL-MSDOS

This is a set of commonly used routines written in optimized assembly language and callable by the user's assembly program. A 'C' callable version is included with the dsp32c-cc-sun4 compiler. The DSP32C-al-sun4 contains the source code for the applications library on cartridge tape and the msdos versions are supplied on floppies.

Arithmetic Routines

_acos calculates arcosine(x) for lxl<0.966		Time: 1.28
_alog10 calculates base 10 anti-log(x) for lxl<38		Time: 1.40
_alog calculates base 2 anti-log(x) for lxl<127		Time: 1.34
_aloge calculates anti-in(x) for lxl<88		Time: 1.40
_asin calculates arcsine(x) for lxl<0.966		Time: 1.23
_atan calculates arctangent(x) for lxl<0		Time: 2.15
_cos calculates cosine(x) for lxl<pi/2		Time: 1.26
_div calculates N/D (divide)		Time: 1.30
_divf faster but less accurate divide		Time: 0.95
_dsp32 converts an array of N numbers in IEEE floating-point format to the internal DSP format.		Time: 0.4+0.10N
_dsp32c same as _dsp32 if N<2049.		Time: .5+.05N
_gran generates Gaussian random number.		Time: 7.93
_ieee32 converts an array of N numbers in DSP internal floating-point format to IEEE format.		Time: .45+.1N
_ieee32c same as _ieee32 if N<2049.		Time: .55+.05N
_inv calculates the inverse of a number.		Time: 1.14
_invf fast inverse routine.		Time: 0.84
_invsqr calculates inverse square root.		Time: 1.83
_log10 calculates base 10 log(x) for lxl<0		Time: 1.71
_log2 calculates base 2 log(x) for lxl<0		Time: 1.66
_loge calculates base e In(x) for lxl<0		Time: 1.71
_ran generates random numbers uniformly distributed between 0.0 and 1.0 with a period of 65536.		Time: 1.03
_ran24 same as _ran but with a period of 16M		Time: 1.03
_sin calculates sine(x) for lxl<pi/2		Time: 1.03
_sqr calculates square root of x for x<0 (accurate to 21 bits).		Time: 1.86
_sqrtq faster but less accurate than _sqr.		Time: 1.3
_tan calculates tangent(x) for lxl<pi/4		Time: 1.76
_xtoy calculates x for x<0.		Time: 2.75
Matrix Routines
_mat2x2 Multiply two 2x2 matrices.		Time: 0.9
_mat3x3 Multiply two 3x3 matrices.		Time: 2.00
_mat4x1 Multiply two 4x4 matrix by 4x1.		Time: 1.3
_mat4x1f Larger but faster version.		Time: 1.25
_mat4x4 Multiply two 4x4 matrices.		Time: 3.85
_mat5x5 Multiply two 5x5 matrices.		Time: 6.9
_matin1 Calculate inverse of a 2x2 matrix.		Time: 1.86
_matinv Calculate inverse of a NxN matrix.		Time for 6x6: 103
_matinf Faster but less accurate than _matinv.		Time for 6x6: 79
_matmul Multiply MxN matrix by NxP.		Time for 5x5: 15.5
Graphics/Imaging Routines
_grey Converts an array of color pixels represented by bits per pixel (5 bits per color) to an array of pixels with a 5-bit grey scale.		Time for N.pixels: .45 + 5.06N
_histeq Implements the histogram equalizatrion algorithm on a grey scale image with each pixel represented by a 5-bit value.		Time for N.pixels: 17.31 + 1.76N
Filter Routines
_fir calculates an fir filter.		Time for length: 1.74
_fir5 fir filter with length multiple of 5.		Time for length: 1.74
_firc complex FIR filter.		Time for length: 4.83
_iir calculates an IIR filter.		Time for length: 5.21
_iir2 calculates a 2-section and 4-multiply per section IIR filter.		Time: 1.01
_iir3 calculates a 3-section and 4-multiply per section IIR filter.		Time: 1.24
_iir4 calculates a 4-section and 4-multiply per section IIR filter.		Time: 1.46
_iird calculates an IIR filter with direct form II cascade 2nd order sections with 5 multiplications/section.		Time for 4 sections: 1.78
_iirdN same as _iird with N sections.		Time for 4 sections: 1.63
_iirt calculates an IIR filter with direct form I cascade 2nd order sections with 5 multiplications/section.		Time for 4 sections: 1.76
_iirt1 calculates a single 5-multiplier section IIR filter.		Time: .79
_iirt2 calculates a 2-section 5-multiplier IIR filter.		Time: 1.06
_iirt3 calculates a 3-section 5-multiplier IIR filter.		Time: 1.34
_iirt4 calculates a 4-section 5-multiplier IIR filter.		Time: 1.61
_osc returns a sample value from a digital sinusoidal osc		Time: 0.51
_oscN returns N samples from a digital sinusoidal osc. _oscamp returns an estimate of the square amplitude of the osc from its state variables.		Time: 1.55
_oscinit initializes the digital sinusoidal oscillator.		Time: 2.1
_tone returns a sample value from a tone-pair.		Time: 0.68
_toneN returns N samples from a tone-pair. _toneamp returns an estimate of the square amplitude of a pair of oscillators.		Time: 2.95
_toneini initializes the pair of oscillators.		Time: 4.35
Adaptive Filter Routines
_Ims Implements a real adaptive FIR filter using the least-mean-square algorithm without tap leakage.		Time for length 20: 3.28
_Imsc Implements a complex adaptive FIR.		Time for length 20: 11.38
_Imsl Implements a real adaptive FIR filter using the leaky least-mean-square algorithm.		Time for length 20: 4.51
FFT Routines
Several Fast Fourier Transform routines handle a combination of real or complex data (in same or separate arrays), from 64 point up to 16K.
Time for a 1024 pt. complex FFT: 2.11ms Time for a 1024 pt. real FFT: 1.31ms
_chammo multiply an array of complex data by the Hamming window.
_channo multiply an array of complex data by the Hanning window.
_hamm multiply an array of real data by the Hamming window.
_hann multiply an array of complex data by the Hanning window.
_ifft inverse complex FFT.		Time for 1024 pt.: 2.84ms

Sampling Rate Conversions

_dnsamp Used when ratio of new sampling freq. to the original freq. is between 0.5 and 1.0.
_dnsamp3 Used when ratio of new sampling freq. to the original freq. is 1/3.
_upsamp Used when ratio of new sampling freq. to the original freq. is greater than 1.
_upsamp3 Used when ratio of new sampling freq. to the original freq. is 3.
The execution times above are in microseconds (unless specified otherwise) for the 80 MHz dsp32c. Multiply by 1.6 for execution time of 50 MHz DSP32c.