Hitachi, Ltd.

Hitachi Denshi Kabushiki Kaisha

Shizuo Sugiyama

Masahito Kobayashi

Yoshimune Hagiwara

Takashi Akazawa

Narimichi Maeda

Yasuhiro Kita

Yuzo Kita

Osamu Yumoto

A signal processor having a wide dynamic range and which can process both data in the fixed point representation and data in the floating point representation by the use of a single floating-point arithmetic circuit is capable of processing digital signals, such as voice signals, at high speed and in real time. In addition, this signal processor includes an arithmetic logic unit for floating point data and/or fixed point data in which there is selectively provided a first pair of first and second floating point data which are to be subjected to an arithmetic operation, or a second pair of data consisting of third floating (fixed) point data which is to be converted to fixed (floating) point data and fourth floating point data which is a reference data for the conversion. If the first pair is selected the first and second pair of floating point data are subjected to the arithmetic operation. If the second pair is selected, the conversion of the fixed (floating) point data to floating (fixed) point data is effected by normalizing the third data with the value of the exponent part of the fourth data.

Arithmetic logic unit for floating point data and/or fixed point data

In an adder for floating point data, two floating point data are adjusted so that the exponent parts have the same value and the resulting adjusted mantissa parts are added. A first shift signal is generated on the basis of the result of the added mantissa parts and having a value necessary for normalization of the addition result, and a second shift signal is generated having a value equal to the difference between the adjusted exponent part of the floating point data and a minimum value predetermined for an exponent of any floating point data at which underflow occurs. The result of addition of the adjusted mantissa parts is shifted on the basis of said second shift signal or said first shift signal depending on whether or not an underflow occurs.

Adder for floating point data suitable for digital signal processing

According to the present invention, a normalized floating point type multiplier circuit including a multiplier for mantissa's multiplication and an adder for exponent's addition is equipped with both a detector circuit for detecting over- and under-flows and a compensation circuit for compensating the output of said multiplier without any programming, when the over- and under-flows are detected, so that the multiplying speed can be improved.

Floating point type multiplier circuit with compensation for over-flow

A signal processor having a wide dynamic range and which can process both data in the fixed point representation and data in the floating point representation by the use of a single floating-point arithmetic circuit is capable of processing digital signals, such as voice signals, at high speed and in real time. In addition, this signal processor is capable of executing data input/output operations with an external circuit in the data format of the fixed point representation and of performing internal operations in the floating point representation format. Further, conversion of an operational result from fixed point representation to floating point representation, and vice versa, can be performed internally in accordance with program instruction.

Growing community of inventors

Kodaira, Japan

Shizuo Sugiyama