GB1037737A - Improvements in or relating to computer systems - Google Patents

Abstract

1,037,737. Calculating transforms of functions. AMPEX CORPORATION. May 4, 1965 [July 1, 1964], No. 18705/65. Heading G4A. An input function f(t) is sampled at regular intervals and the sampled data multiplied by the simultaneous value of a trigonometric function generated so that its frequency is proportional to time, the products being successively added in a storage device, from which at the end of the sampling period, the transform of the input function is available. Fourier transform.-The input function which may be supplied by transducer elements or a magnetic tape recorder is sampled when the sample means 20 is enabled by control means 90 (Fig. 3), the result of the sample being fed to analog multiplier means 30, 32. Function generator means producing a sine and cosine function may comprise three generating means 41-43 (Fig. 4a, not shown) emitting signals of frequency 50 c.p.s., 100 c.p.s. and 200 c.p.s. respectively connected to frequency adder means so that, under the control of logic means 47, frequencies of 50-350 c.p.s. may be applied to the multiplier means, the sine function to multiplier 30 and the cosine to multiplier 32. Alternatively the generating means may comprise a plurality of bi-stable devices 54-57 (Fig. 4b, not shown) input pulses from control means 90 being supplied at the rate of m.p.s. to the first device 54 m to give an output ofper second to the second 2 device &c. Each is connected to a low-pass filter 58-61 which extracts the first harmonic from the square wave output of the bi-stable devices. This is applied under the control of a binary counter 71 to single sideband modulators 67-70 and phase shifters 72a-72d which shift the sine waveforms through 90 degrees to transform them into cosine waveforms. When the counter 71 is in its zero position, zero and D.C. voltages are applied to the sine and cosine output terminals respectively. When it is registering a negative value an inverter 74 is enabled to shift the sine waveform through 180 degrees. The output from the multipliers 30, 32 is fed via analogue to digital converter means 75 to digital adder means 80, 82 respectively which are connected to the sections 96, 98 of a core memory, the number stored in each section being fed back to the adders 80, 82 simultaneously with each sample/frequency product so that, at the end of the sample, the real and imaginary parts of the Fourier transform may be supplied from sections 94, 98 respectively to magnetic tape systems, data blotters, cathoderay tubes, punched cards or printers. Fourier transform of a complex function.- The Fourier transform of a complex function is similarly calculated by applying the real portion to sample means 20a (Fig. 6, not shown) and the imaginary portion to sample means 20b. Each portion is then multiplied by the sine and cosine generated functions and the appropriate products added together at 110, 120, the results being fed to a memory 100. Power density spectra.-The power density spectra is calculated by feeding both the real and imaginary parts of the Fourier transform from core memory 96 twice to multiplier means and then adding the resultant products, the output of which is in analogue form. This output is fed to an analogue-digital converter before being entered back into the memory. Cross spectra.-The cross spectra is similarly found by providing four multipliers calculating the products of the real and imaginary parts of two Fourier transforms and cross products. Auto- or cross-correlation.-In this case the power density spectra is first calculated and transferred from the core memory means via a digital-analogue converter to sample means, the result of the sample being multiplied a cosine generated function for autocorrelation and additionally by a sine generated function for cross correlation, the products being accumulated in the memory means.