Multiplication is an essential image processing operation commonly implemented in hardware DSP co... more Multiplication is an essential image processing operation commonly implemented in hardware DSP cores. To improve DSP cores’ area, speed, or energy efficiency, we can approximate multiplication. We present an approximate multiplier that generates two partial products using hybrid radix-4 and logarithmic encoding of the input operands. It uses the exact radix-4 encoding to generate the partial product from the three most significant bits and the logarithmic approximation with mantissa trimming to approximate the partial product from the remaining least-significant bits. The proposed multiplier fills the gap between highly accurate approximate non-logarithmic multipliers with a complex design and less accurate approximate logarithmic multipliers with a more straightforward design. We evaluated the multiplier’s efficiency in terms of error, energy (power-delay-product) and area utilisation using NanGate 45 nm. The experimental results show that the proposed multiplier exhibits good area...
Compression of convolutional neural networks: A short survey
2018 17th International Symposium INFOTEH-JAHORINA (INFOTEH), 2018
Nowadays, convolutional neural networks (CNN) are considered as the state-of-the-art algorithms f... more Nowadays, convolutional neural networks (CNN) are considered as the state-of-the-art algorithms for various tasks, especially for image classification and recognition. Because of that, more and more attention is aimed towards implementation of CNNs on embedded systems. Main obstacles for implementing CNNs on embedded systems are their large model size and large number of operations needed for inference. In order to surpass these obstacles, algorithms for CNN compression tend to lower model size and number of operations needed for inference. In this paper we review the state-of-the-art in CNN compression. To this end we divided all approaches for CNN compression into three groups: precision reduction, network pruning and design of compact network architectures. After presenting the main approaches in each group we conclude that the future CNN compression algorithms should be co-designed with hardware which will process deep learning algorithms.
This paper presents a new A-weighting filter’s design and explores the potential of using approxi... more This paper presents a new A-weighting filter’s design and explores the potential of using approximate multiplication for low-power digital A-weighting filter implementation. It presents a thorough analysis of the effects of approximate multiplication, coefficient quantization, the order of first-order sections in the filter’s cascade, and zero-pole pairings on the frequency response of the digital A-weighting filter. The proposed A-weighting filter was implemented as a sixth-order IIR filter using approximate odd radix-4 multipliers. The proposed filter was synthesized (Verilog to GDS) using the Nangate45 cell library, and MATLAB simulations were performed to verify the designed filter’s magnitude response and performance. Synthesis results indicate that the proposed design achieves nearly 70% reduction in energy (power-delay product) with a negligible deviation of the frequency response from the floating-point implementation. Experiments on acoustic noise suggest that the proposed ...
IEEE Transactions on Circuits and Systems I: Regular Papers, 2021
We present an approximate logarithmic multiplier with two-stage operand trimming, which prioritis... more We present an approximate logarithmic multiplier with two-stage operand trimming, which prioritises area and energy consumption while retains acceptable accuracy. The multiplier trims the least significant parts of input operands in the first stage and the mantissas of the obtained operands' approximations in the second stage. We evaluated the multiplier's efficiency in terms of error, energy, and area utilisation using NanGate 45nm. The experimental results show that the proposed multiplier exhibits smaller area utilisation and energy consumption than the state-of-the-art designs and that it behaves well in image processing and image classification with convolutional neural networks.
Undergraduate Topics in Computer Science (UTiCS) delivers high-quality instructional content for ... more Undergraduate Topics in Computer Science (UTiCS) delivers high-quality instructional content for undergraduates studying in all areas of computing and information science. From core foundational and theoretical material to final-year topics and applications, UTiCS books take a fresh, concise, and modern approach and are ideal for self-study or for a one-or two-semester course. The texts are all authored by established experts in their fields, reviewed by an international advisory board, and contain numerous examples and problems. Many include fully worked solutions.
One important aspect when choosing a Bluetooth Low Energy (BLE) solution is to analyze its energy... more One important aspect when choosing a Bluetooth Low Energy (BLE) solution is to analyze its energy consumption for various connection parameters and desired throughput to build an optimal low-power Internet-of-Things (IoT) application and to extend the battery life. In this paper, energy consumption and data throughput for various BLE versions are studied. We have tested the effect of connection interval on the throughput and compared power efficiency relating to throughput for various BLE versions and different transactions. The presented results reveal that shorter connection intervals increase throughput for read/write transactions, but that is not the case for the notify and read/write without response transactions. Furthermore, for each BLE version, the energy consumption is mainly dependable on the data volume. The obtained results provide a design guideline for implementing an optimal BLE IoT application.
To reduce the power dissipation in adaptive control systems, we propose replacing the exact arith... more To reduce the power dissipation in adaptive control systems, we propose replacing the exact arithmetic hardware units with approximate ones. As a case study, an adaptive control system for object tracking based on the Kalman filter is implemented in FPGA. A thorough analysis of the Kalman filter's circuitry for real-world object tracks acquired by an aviation radar system proved that adaptive control systems can successfully compensate for the calculation errors introduced by the approximate arithmetic units. The main contributions of this paper are that the introduction of the approximate arithmetic circuits to the adaptive control system (1) preserves the required accuracy and (2) significantly reduces the power dissipation and the size of the adaptive system's circuitry.
This paper presents the vector C (VC) language, which is designed for the multimedia extensions i... more This paper presents the vector C (VC) language, which is designed for the multimedia extensions included in all modern microprocessors. The paper discusses the language syntax, the implementation of its compiler and its use in developing multimedia applications. The goal was to provide programmers with the most natural way of using multi-media processing facilities in the C language. The VC language has been used to develop some of the most frequently used multimedia kernels. The experiments on these scientific and multimedia applications have yielded good performance improvements
There is an increasing need for fast and efficient algorithms for the automatic analysis of remot... more There is an increasing need for fast and efficient algorithms for the automatic analysis of remote-sensing images. In this paper we address the implementation of the semantic classification of aerial images with general-purpose graphicsprocessing units (GPGPUs). We propose the calculation of a local Gabor-based structural texture descriptor and a structural texture similarity metric combined with a nearest-neighbor classifier and image-to-class similarity on CUDA supported graphics-processing units. We first present the algorithm and then describe the GPU implementation and optimization with the CUDA programming model. We then evaluate the results of the algorithm on a dataset of aerial images and present the execution times for the sequential and parallel implementations of the whole algorithm as well as measurements only for the selected steps of the algorithm. We show that the algorithms for the image classification can be effectively implemented on the GPUs. In our case, the presented algorithm is around 39 times faster on the Tesla C1060 unit than on the Core i5 650 CPU, while keeping the same success rate of classification.
In this article we present the use of the logarithmic number system (LNS) to implement fixed-poin... more In this article we present the use of the logarithmic number system (LNS) to implement fixed-point multiplication and division. LNS has recently attracted the interest of researchers for its low-power properties. The reduction of power dissipation in LNS arises from the simplification of basic arithmetic operations. In this paper we give a survey of the recently proposed digital circuits for logarithm and anti-logarithm conversion and multiplication and division in LNS. We also compare these methods in terms of accuracy, area, time and power. Finally, we give an overview of the real world applications that benefit form the use of LNS arithmetic.
Edge computing brings artificial intelligence algorithms and graphics processing units closer to ... more Edge computing brings artificial intelligence algorithms and graphics processing units closer to data sources, making autonomy and energy-efficient processing vital for their design. Approximate computing has emerged as a popular strategy for energy-efficient circuit design, where the challenge is to achieve the best tradeoff between design efficiency and accuracy. The essential operation in artificial intelligence algorithms is the general matrix multiplication (GEMM) operation comprised of matrix multiplication and accumulation. This paper presents an approximate general matrix multiplication (AGEMM) unit that employs approximate multipliers to perform matrix–matrix operations on four-by-four matrices given in sixteen-bit signed fixed-point format. The synthesis of the proposed AGEMM unit to the 45 nm Nangate Open Cell Library revealed that it consumed only up to 36% of the area and 25% of the energy required by the exact general matrix multiplication unit. The AGEMM unit is ideal...
Accurate Indoor Sound Level Measurement on a Low-Power and Low-Cost Wireless Sensor Node
Wireless sensor networks can provide a cheap and flexible infrastructure to support the measureme... more Wireless sensor networks can provide a cheap and flexible infrastructure to support the measurement of noise pollution. However, the processing of the gathered data is challenging to implement on resource-constrained nodes, because each node has its own limited power supply, low-performance and low-power micro-controller unit and other limited processing resources, as well as limited amount of memory. We propose a sensor node for monitoring of indoor ambient noise. The sensor node is based on a hardware platform with limited computational resources and utilizes a number of simplifications to approximate more complex and costly signal processing stage. Furthermore, to reduce the communication between the sensor node and a sink node, as well as the power consumed by the IEEE 802.15.4 (ZigBee) transceiver, we perform digital A-weighting filtering and non-calibrated calculation of the sound pressure level on the node. According to experimental results, the proposed sound level meter can...
Multiplication is an essential image processing operation commonly implemented in hardware DSP co... more Multiplication is an essential image processing operation commonly implemented in hardware DSP cores. To improve DSP cores’ area, speed, or energy efficiency, we can approximate multiplication. We present an approximate multiplier that generates two partial products using hybrid radix-4 and logarithmic encoding of the input operands. It uses the exact radix-4 encoding to generate the partial product from the three most significant bits and the logarithmic approximation with mantissa trimming to approximate the partial product from the remaining least-significant bits. The proposed multiplier fills the gap between highly accurate approximate non-logarithmic multipliers with a complex design and less accurate approximate logarithmic multipliers with a more straightforward design. We evaluated the multiplier’s efficiency in terms of error, energy (power-delay-product) and area utilisation using NanGate 45 nm. The experimental results show that the proposed multiplier exhibits good area...
Compression of convolutional neural networks: A short survey
2018 17th International Symposium INFOTEH-JAHORINA (INFOTEH), 2018
Nowadays, convolutional neural networks (CNN) are considered as the state-of-the-art algorithms f... more Nowadays, convolutional neural networks (CNN) are considered as the state-of-the-art algorithms for various tasks, especially for image classification and recognition. Because of that, more and more attention is aimed towards implementation of CNNs on embedded systems. Main obstacles for implementing CNNs on embedded systems are their large model size and large number of operations needed for inference. In order to surpass these obstacles, algorithms for CNN compression tend to lower model size and number of operations needed for inference. In this paper we review the state-of-the-art in CNN compression. To this end we divided all approaches for CNN compression into three groups: precision reduction, network pruning and design of compact network architectures. After presenting the main approaches in each group we conclude that the future CNN compression algorithms should be co-designed with hardware which will process deep learning algorithms.
This paper presents a new A-weighting filter’s design and explores the potential of using approxi... more This paper presents a new A-weighting filter’s design and explores the potential of using approximate multiplication for low-power digital A-weighting filter implementation. It presents a thorough analysis of the effects of approximate multiplication, coefficient quantization, the order of first-order sections in the filter’s cascade, and zero-pole pairings on the frequency response of the digital A-weighting filter. The proposed A-weighting filter was implemented as a sixth-order IIR filter using approximate odd radix-4 multipliers. The proposed filter was synthesized (Verilog to GDS) using the Nangate45 cell library, and MATLAB simulations were performed to verify the designed filter’s magnitude response and performance. Synthesis results indicate that the proposed design achieves nearly 70% reduction in energy (power-delay product) with a negligible deviation of the frequency response from the floating-point implementation. Experiments on acoustic noise suggest that the proposed ...
IEEE Transactions on Circuits and Systems I: Regular Papers, 2021
We present an approximate logarithmic multiplier with two-stage operand trimming, which prioritis... more We present an approximate logarithmic multiplier with two-stage operand trimming, which prioritises area and energy consumption while retains acceptable accuracy. The multiplier trims the least significant parts of input operands in the first stage and the mantissas of the obtained operands' approximations in the second stage. We evaluated the multiplier's efficiency in terms of error, energy, and area utilisation using NanGate 45nm. The experimental results show that the proposed multiplier exhibits smaller area utilisation and energy consumption than the state-of-the-art designs and that it behaves well in image processing and image classification with convolutional neural networks.
Undergraduate Topics in Computer Science (UTiCS) delivers high-quality instructional content for ... more Undergraduate Topics in Computer Science (UTiCS) delivers high-quality instructional content for undergraduates studying in all areas of computing and information science. From core foundational and theoretical material to final-year topics and applications, UTiCS books take a fresh, concise, and modern approach and are ideal for self-study or for a one-or two-semester course. The texts are all authored by established experts in their fields, reviewed by an international advisory board, and contain numerous examples and problems. Many include fully worked solutions.
One important aspect when choosing a Bluetooth Low Energy (BLE) solution is to analyze its energy... more One important aspect when choosing a Bluetooth Low Energy (BLE) solution is to analyze its energy consumption for various connection parameters and desired throughput to build an optimal low-power Internet-of-Things (IoT) application and to extend the battery life. In this paper, energy consumption and data throughput for various BLE versions are studied. We have tested the effect of connection interval on the throughput and compared power efficiency relating to throughput for various BLE versions and different transactions. The presented results reveal that shorter connection intervals increase throughput for read/write transactions, but that is not the case for the notify and read/write without response transactions. Furthermore, for each BLE version, the energy consumption is mainly dependable on the data volume. The obtained results provide a design guideline for implementing an optimal BLE IoT application.
To reduce the power dissipation in adaptive control systems, we propose replacing the exact arith... more To reduce the power dissipation in adaptive control systems, we propose replacing the exact arithmetic hardware units with approximate ones. As a case study, an adaptive control system for object tracking based on the Kalman filter is implemented in FPGA. A thorough analysis of the Kalman filter's circuitry for real-world object tracks acquired by an aviation radar system proved that adaptive control systems can successfully compensate for the calculation errors introduced by the approximate arithmetic units. The main contributions of this paper are that the introduction of the approximate arithmetic circuits to the adaptive control system (1) preserves the required accuracy and (2) significantly reduces the power dissipation and the size of the adaptive system's circuitry.
This paper presents the vector C (VC) language, which is designed for the multimedia extensions i... more This paper presents the vector C (VC) language, which is designed for the multimedia extensions included in all modern microprocessors. The paper discusses the language syntax, the implementation of its compiler and its use in developing multimedia applications. The goal was to provide programmers with the most natural way of using multi-media processing facilities in the C language. The VC language has been used to develop some of the most frequently used multimedia kernels. The experiments on these scientific and multimedia applications have yielded good performance improvements
There is an increasing need for fast and efficient algorithms for the automatic analysis of remot... more There is an increasing need for fast and efficient algorithms for the automatic analysis of remote-sensing images. In this paper we address the implementation of the semantic classification of aerial images with general-purpose graphicsprocessing units (GPGPUs). We propose the calculation of a local Gabor-based structural texture descriptor and a structural texture similarity metric combined with a nearest-neighbor classifier and image-to-class similarity on CUDA supported graphics-processing units. We first present the algorithm and then describe the GPU implementation and optimization with the CUDA programming model. We then evaluate the results of the algorithm on a dataset of aerial images and present the execution times for the sequential and parallel implementations of the whole algorithm as well as measurements only for the selected steps of the algorithm. We show that the algorithms for the image classification can be effectively implemented on the GPUs. In our case, the presented algorithm is around 39 times faster on the Tesla C1060 unit than on the Core i5 650 CPU, while keeping the same success rate of classification.
In this article we present the use of the logarithmic number system (LNS) to implement fixed-poin... more In this article we present the use of the logarithmic number system (LNS) to implement fixed-point multiplication and division. LNS has recently attracted the interest of researchers for its low-power properties. The reduction of power dissipation in LNS arises from the simplification of basic arithmetic operations. In this paper we give a survey of the recently proposed digital circuits for logarithm and anti-logarithm conversion and multiplication and division in LNS. We also compare these methods in terms of accuracy, area, time and power. Finally, we give an overview of the real world applications that benefit form the use of LNS arithmetic.
Edge computing brings artificial intelligence algorithms and graphics processing units closer to ... more Edge computing brings artificial intelligence algorithms and graphics processing units closer to data sources, making autonomy and energy-efficient processing vital for their design. Approximate computing has emerged as a popular strategy for energy-efficient circuit design, where the challenge is to achieve the best tradeoff between design efficiency and accuracy. The essential operation in artificial intelligence algorithms is the general matrix multiplication (GEMM) operation comprised of matrix multiplication and accumulation. This paper presents an approximate general matrix multiplication (AGEMM) unit that employs approximate multipliers to perform matrix–matrix operations on four-by-four matrices given in sixteen-bit signed fixed-point format. The synthesis of the proposed AGEMM unit to the 45 nm Nangate Open Cell Library revealed that it consumed only up to 36% of the area and 25% of the energy required by the exact general matrix multiplication unit. The AGEMM unit is ideal...
Accurate Indoor Sound Level Measurement on a Low-Power and Low-Cost Wireless Sensor Node
Wireless sensor networks can provide a cheap and flexible infrastructure to support the measureme... more Wireless sensor networks can provide a cheap and flexible infrastructure to support the measurement of noise pollution. However, the processing of the gathered data is challenging to implement on resource-constrained nodes, because each node has its own limited power supply, low-performance and low-power micro-controller unit and other limited processing resources, as well as limited amount of memory. We propose a sensor node for monitoring of indoor ambient noise. The sensor node is based on a hardware platform with limited computational resources and utilizes a number of simplifications to approximate more complex and costly signal processing stage. Furthermore, to reduce the communication between the sensor node and a sink node, as well as the power consumed by the IEEE 802.15.4 (ZigBee) transceiver, we perform digital A-weighting filtering and non-calibrated calculation of the sound pressure level on the node. According to experimental results, the proposed sound level meter can...
Uploads
Papers by Patricio Bulic