With the introduction of the Fast Fourier Transform (FFT) in the 1960s and the refinement of Infinite Impulse Response (IIR) and Finite Impulse Response (FIR) filter designs, the demand for digital signal processing technology began to grow rapidly. During this period, significant increases in demand across fields such as audio, image, and communications drove the development and application of Digital Signal Processors (DSPs).
In the audio field, Audio DSP technology developed rapidly, significantly improving system sound quality through innovative audio algorithm designs. With the breakthrough progress in artificial intelligence technology and the rise of Neural Processing Units (NPUs), future audio processing technology will develop in the direction of combining DSP and NPU, providing a new direction for the intelligent development of audio processing.
The Awinic Soaring™ DSP launched by Awinic, a leader in digital and analog technology, is a high-performance solution born to meet modern audio processing needs. The term "Soaring" embodies Eastern aesthetics and imagination, originating from the image of the "Eastern Goddess" flying in mythology. "Fei" symbolizes the breakthrough of sound physical limitations for an acoustic leap, while "Tian" represents reaching the highest realm of pure sound quality. The name "Soaring" echoes the "Immortal Algorithm SKTune," carrying the imagery of traditional culture, Eastern romance, and technological breakthrough.
This series will elaborate on the technological evolution of DSP and the corresponding product evolution of Awinic Soaring™ DSP from the following aspects:
Evolution and Design of DSP Technology
Evolution of Awinic Soaring™ DSP Products
Introduction to Awinic Soaring™ DSP's Characteristic Algorithms
In recent years, DSPs (Digital Signal Processors) have continuously evolved in terms of architecture design, instruction execution, process technology, and algorithm adaptation. Through technological innovation and optimized design, not only has the performance of DSPs been significantly improved, but their power consumption requirements have also been effectively reduced.
DSP architecture is the core of digital signal processing systems, and its design directly affects signal processing efficiency and the overall performance of the system.
Von Neumann Architecture
The Von Neumann architecture is the foundation of modern computers. Currently, the architectures of almost all computer CPUs originate from the Von Neumann architecture. The Von Neumann architecture defines the basic relationship between the CPU and memory. Data transmission between CPUs occurs via a bus, which can transmit either instructions or data. Instructions and data are in the same storage area and use the same bus for transmission. During data processing, since both data and instructions are processed on the same bus, the two must be processed serially. Therefore, basic operations such as instruction fetch and data fetch cannot be performed simultaneously. The basic operations of instructions are illustrated in Figure 1.
Harvard Architecture
The Harvard structure uses different buses for data and instruction transmission, allowing instruction fetch and decode operations to be processed in parallel. For example, while decoding instruction 1, instruction 2 can be fetched simultaneously. This is because data transmission occurs on two separate buses, allowing these two operations to be performed in parallel, greatly enhancing processor performance. DSPs are almost exclusively processors that use the Harvard architecture.
With technological development and higher demands for computation, DSP computation needs to evolve in terms of broadband and vectorization. Instruction processing has gradually shifted from Single Instruction Single Data (SISD) to Single Instruction Multiple Data (SIMD). Below are the differences between these technologies. SIMD technology has evolved and developed across various DSPs or CPUs:
SIMD registers can be divided into 64-bit and 128-bit. Data in a 64-bit register can be divided into: two 32-bit, four 16-bit, or eight 8-bit integer data elements that can be operated on simultaneously in a 64-bit register; data in a 128-bit register can be divided into: four 32-bit, eight 16-bit, or sixteen 8-bit integer data elements that can be operated on simultaneously in a 128-bit register. This SIMD operation mode greatly enhances the computational performance of DSPs/CPUs.
In the subsequent technological evolution of DSP/CPU manufacturers, some adopted the Very Long Instruction Word (VLIW) SIMD mode to improve parallel computing capabilities. For example, Qualcomm's DSP HVX (Hexagon Vector eXtensions) instructions and ARM's NEON instructions significantly enhance parallel computing efficiency by extending vector processing capabilities. Awinic Soaring™ DSP has also conducted in-depth research and practical applications in the field of vector computing to optimize its processing performance and efficiency.
In addition to computational performance, process technology is another key area in DSP technology development. In the 1990s, with the further development of CMOS technology, the power consumption of DSP chips was significantly reduced. The low-power characteristics of CMOS technology enabled DSPs to be more widely used in fields such as speech processing and image processing. As a leader in the DSP industry, Texas Instruments (TI) used 0.18-micron process technology for its C5000 series, while the C6000 series upgraded to 90nm or 65nm process technology. Compared to the C5000, the C6000 showed significant improvements in both power consumption and performance.
The evolution of process technology not only significantly reduced chip area, making it easier to integrate into single-board systems, but also notably lowered power consumption. For example, the power consumption of a mobile high-definition encoder IP using 90nm process technology is about 100 milliwatts, while with 65nm process technology, it can be reduced to 50 milliwatts, saving approximately 50% in power consumption.
In achieving low-power design, besides improvements in process technology, the evolution of other technologies is equally important. Dynamic Voltage and Frequency Scaling (DVFS) is a technology that optimizes power consumption by adjusting operating voltage and frequency in real-time. Under light load conditions, the DSP can significantly reduce power consumption by lowering voltage and frequency. This technology relies on accurate load prediction and fast switching of operating points, achieved by dynamically adjusting based on instruction execution statistics from performance counters.
From the above introduction, it can be seen that current DSP architectures are all based on the Harvard structure, which can effectively improve system efficiency. Meanwhile, the SIMD instruction processing method significantly enhances DSP processing capability from another dimension. Additionally, in the development of DSPs, process technology has an important impact on their performance, with more advanced processes leading to optimizations in power consumption and area.
Awinic, a leader in digital and analog technology in China, has launched the Awinic Soaring™ DSP. Throughout its product evolution, it has consistently committed to absorbing advanced technological achievements, combining them with its own needs and product characteristics for targeted optimization and innovation, ultimately creating highly adaptable product solutions. The following will elaborate in detail on the current product iterations and evolution of Awinic Soaring™ DSP.
The figure below shows the changes in main features and key parameters from the first-generation Awinic Soaring™ DSP product to the latest generation.
As can be seen from Figure 4, from the first-generation 0.18μm process to the 55nm process, computing power increased by 4 times, and power consumption reduced by 74%.
The first-generation Awinic Soaring™ DSP product adopted 0.18μm process technology, achieving the integration of Audio DSP functions and algorithms. However, limited by the characteristics of the 0.18μm process, the chip area of this generation was relatively large. The second-generation product introduced more advanced 90nm process technology and successfully applied it in the AW88166 product. Through process upgrading, not only was power consumption significantly reduced, but algorithm modules such as virtual bass and large volume were added, further enhancing product functionality and performance.
In the latest DSP products, Awinic Soaring™ DSP adopts 55nm process technology, which comprehensively improves product performance, power consumption, and area, further enhancing its market competitiveness.
The computational performance of Awinic Soaring™ DSP has undergone significant improvement, gradually upgrading from 40 MIPS in the first generation to 200 MIPS in the latest generation DSP 5.0, demonstrating continuous optimization and enhancement in computing capability. This improvement is not only reflected in the significant increase in processing speed but also provides stronger support for algorithm complexity and functional expansion.
In the latest generation of DSP algorithms, Awinic Soaring™ DSP introduces AI-based deep learning technology, further expanding its application scope and performance. AI algorithms are applied in areas such as sound field surround optimization, low-frequency enhancement, and loudness improvement. Through intelligent sound processing, users are provided with an immersive sound experience, bringing sound effects close to the level of professional concert halls. The introduction of this technology not only enhances the precision and richness of audio processing but also provides users with higher-quality auditory enjoyment.
Awinic, a leader in digital and analog technology, has formed a rich series of audio products through continuous technological innovation and product development of Awinic Soaring™ DSP. In the new generation of products, the introduction of AI technology achieves the creation of immersive and 3D surround sound effects, significantly improving user experience.
High-quality DSP products need to carry excellent DSP algorithms to reflect the core value of the product. The characteristic algorithms of Awinic Soaring™ DSP play a key role in enhancing product performance and optimizing functions. At the same time, this article will also look forward to the future development trends of DSP technology and explore its potential technological evolution directions.
Awinic has further upgraded the protection requirements under low battery scenarios, creating a Low Voltage Protection (LTP) algorithm with microsecond-level response speed, multi-level debugging, and various debugging methods. This allows the power amplifier to still work normally without noise even at lower power supply voltages.
As shown in the figure above, when the power supply voltage drops from 3.8V to 2.7V, AW88166 can still output music signals without distortion and noise (the low voltage protection algorithm activates, and the overall amplitude decreases slightly), and it still produces sound normally at 2.7V low voltage; audio power amplifiers without low voltage protection will enter an abnormal working state and output no sound under a 2.7V power supply voltage.
Because the LTP algorithm is integrated in the chip's DSP, the algorithm's response speed can reach the microsecond level (the response time of other low voltage protection solutions without DSP is much greater than 10ms). The battery voltage is obtained through the internal Vsensor, the chip temperature is obtained through the internal Temp Sensor, and the external temperature is obtained through I²C. Thus, the LTP algorithm module can simultaneously monitor the battery voltage, external temperature, and chip temperature. According to different temperatures and voltages, it can dynamically configure the chip's maximum drawn current Ipeak from the battery, preventing the battery voltage from being drawn too low in low-temperature and low-battery scenarios, which could cause the chip to trigger UVLO and enter an abnormal working state or cause the mobile device to force shutdown. Meanwhile, when Ipeak decreases and the chip's maximum output capability declines, the LTP module can control the chip's Volume and Vmax to ensure that the maximum output waveform is not clipped and there is no noise.
The LTP algorithm is not only used in AW88166 but also in subsequent new Awinic Soaring™ DSP products such as AW88394/AW88399/AW85180S.
Traditional piano noise suppression methods use tuning techniques like EQ/DRC to suppress noise, which affects signals that do not produce noise. The piano noise suppression algorithm from Awinic, a leader in digital and analog technology, can identify and suppress noise through machine learning:
Identify noise segments, eliminate noise while ensuring other parts of the music are unaffected;
Only suppress large signals that produce noise, with little or no processing for small and medium signals, resulting in minimal subjective effect impact and no additional cost.
The suppression gain and noise waveform can be represented as follows:
When machine learning identifies piano noise that needs suppression, it checks the signal amplitude and provides the corresponding suppression gain. When the noise signal is larger, the suppression amplitude is greater; when the noise signal is smaller, the suppression amplitude is smaller.
The following shows the effect diagrams of different processing methods for noise suppression. The left diagram shows the effect with suppression:
The effects of traditional methods and Awinic's noise suppression method are basically similar. Figure 8 shows that when suppression processing is not performed, traditional methods suppress the signal, while Awinic's method has little impact on non-noise processing.
Awinic Soaring™ DSP products such as AW85815/AW85180 support the piano noise suppression algorithm.
With the diversification of personal entertainment scenarios and the ecological integration of smart terminals, the limitations of traditional two-channel systems in terms of spatial sense restoration, sound source localization accuracy, and dynamic range performance are becoming increasingly apparent. Virtual surround sound fields expand sound from a two-dimensional plane to a three-dimensional space, achieving immersive auditory experiences such as raindrops falling from overhead or bullets whizzing past from the side and rear. Breakthroughs in AI technology directly promote personal entertainment scenarios.
Through AI recognition, separate vocal signals, accompaniment signals, and ambient sound signals in the left and right channels;
Perform gain processing on vocals, accompaniment, and ambient sound respectively to achieve effects like vocal enhancement;
Use head-related transfer functions to perform binaural rendering of signals at different virtual sound source positions;
Break through the sound field limitations imposed by physical speaker positions to create realistic 5.1-channel surround sound effects.
While the sound field is widened, virtual surround sound from the side and rear can be clearly heard, as shown in Figure 10.
The AI sound field surround algorithm function is supported in the Awinic Soaring™ DSP product AW88188.
With the rapid development of artificial intelligence technology, the application scope of AI algorithms continues to expand, and their computational demands increase significantly. In the field of edge devices, the application of AI algorithms is gradually becoming popular, such as in smart speakers, smart home devices, and wearable devices. These devices achieve functions like voice interaction, environmental perception, and health monitoring through localized processing. However, edge devices are limited in computing capacity due to processor size constraints, facing challenges such as limited computational resources, power consumption constraints, and high real-time requirements. How to efficiently implement AI algorithms on edge devices has become an important issue for current Digital Signal Processors (DSPs).
As specialized processors for signal processing, DSPs are already widely used in fields such as audio and image. However, with the increase in intelligent demands, DSPs also need to deeply integrate existing signal processing technologies with AI algorithms to meet more complex processing needs.
To meet the high computational demands of AI algorithms on edge devices, Neural Processing Units (NPUs) have emerged. NPUs, as hardware accelerators specifically designed for neural network computing, can significantly improve the operating efficiency of AI algorithms and become a key technology to address AI computing needs on the edge. Through the combination of NPU and DSP, not only can the operating efficiency and power consumption performance of AI algorithms be optimized, but it also provides important support for performance improvement of edge intelligent devices.
With advantages such as high efficiency, low power consumption, real-time response, and dedicated hardware acceleration, NPUs are becoming key hardware components in AI applications. Their performance is particularly outstanding in scenarios such as edge computing, real-time data analysis, and terminal AI. Awinic, a leader in digital and analog technology in China, will combine its own market demands, technological accumulation, and cost control factors to design more NPU products that meet performance requirements and have market competitiveness.
Overall, power amplifier DSPs still hold an important position. Although platform computing power has significantly improved, the demand for computing power is endless. Power amplifier DSPs can utilize their own computational advantages to supplement application scenarios where platform computing power is insufficient. Meanwhile, power amplifier DSPs have significant advantages in real-time processing and can combine with speakers for timely processing, thus achieving the best protection and sound effects.
This series overall reviews DSP architecture and technology, Awinic Soaring™ DSP product iterations and core algorithms, and provides an in-depth outlook on the future development trends of DSP technology, especially NPU technology and rich AI application demands. Awinic, a leader in digital and analog technology, always closely focuses on DSP technology to carry out innovative work, committed to adopting advanced DSP technology to develop products with the best effects.
沪ICP备10011954号 |
沪公网安备 31011202011955号
EN