Silencions ANC Solution

Silencions s|ANC|

The Silencions ANC solution is new generation of active systems for active noise cancelation. Based on the knowledge and experience of an interdisciplinary team of scientists and engineers in the fields of acoustics, algorithmics and mechatronics, an innovative system was developed based on machine learning for the harsh environmental conditions.

how does the Silencions S|ANC| system work?

THE POWER OF ANC

The use of machine learning and artificial intelligence make this solution unique. The design of an active noise cancellation system is based on the principle of generating an acoustic wave in an inverted phase to phase generated by a source of noise using the machine learning technique. Thanks to the innovative approach to the problem of noise reduction, this innovation stands out from the products present on the market with its unique features, such as:

the use of machine learning technology,
high efficiency, including higher frequencies,
robust design for harsh environmental conditions.

Thanks to the above features, the developed solution can be used in various conditions.

Active Noise Control Solution software

The Silencions system for active noise reduction is improved by using machine learning algorithms. The basis of the system is the use of linear processing algorithms, such as finite impulse response (FIR) filters and their extension with an adaptive algorithm (adaptive filters), i.e. those in which the filter weights are continuously updated. The adaptive filters are considered in the network to improve ANC performance. A significant advantage of the developed solution is the signal delay reduction by the analog-digital processing system. Initially assumed delay value has been significantly reduced from the initial 200 ms to a few ms, both through hardware work and the developed software component. Developed system involves the creation of a solution based on machine learning algorithms for the task of sound wave prediction.

The research was devoted to the development of algorithms for prediction based on neural networks and the creation of our architecture, which will allow to train a model appropriate for the characteristics of the sound signal and its prediction. The work on modeling the architectures of neural networks allowed to select a network consisting of 4 two-dimensional convolutional layers, interlaced with Leaky ReLU activation functions and a recurrent LSTM layer at the end of the network. The input to the dedicated neural network is the sound in the frequency domain, while the output from the network is transformed in the last phase of processing.

Significant work also focused on development of an algorithm accumulating training sets. The main challenge during this work was to design a scheme of operation and communication between individual components of the ANC system and the designed algorithm. The system is able to enlarge the database of signals on which it was trained, so as to increase its effectiveness in new work environments or in environments with a rapidly changing signal. The designed solution works in two modes, learning and operation. Both solutions are based on FxLMS type adaptive filters.

Data collected can be used to optimize the system in many ways. One of them is to use the data to predict the weights of adaptive filters by a neural network. This allows to avoid the long time of the system convergence to optimal weights in the case of sudden changes in the signal course.
A big advantage of such a solution is the fact that the developed ANC system can potentially operate stably without an error microphone, which in a traditional ANC system is placed in the place of the target expected muting. In practice, the location of the error microphone may be impossible to implement due to the need to ensure appropriate room comfort, including privacy and confidentiality, e.g. placing a microphone in a room for adaptive ventilation muting.

Active Noise Control Solution hardware

The device housing was designed specifically for the developed board. Particular emphasis was placed on adapting the housing to the final version of the ANC electronic system. This forced a complete redesign of the electronics housing for any demanding operational needs. Prototypes of individual elements and entire details were manufactured in accordance with the rapid prototyping methodology and were largely based on the use of additive technologies, including 3D printing method LPBF. Internal tests showed compliance with IP34 and IK06 standards, which allowed for the commencement of tests in an external certification body. Fig. 1 below is an assembly drawing of the housing design and in Fig. 2 a photo showing its final construction with all assembly elements.

As part of the pre-implementation work carried out at this stage, the housing of the designed system was prepared for testing and certification. A competent research and certification laboratory was selected to conduct the necessary tests. These tests were carried out with a positive result, confirming that the system housing meets the requirements of the IK06 standard, which indicates its impact resistance, and the internal protection standard. Based on the tests carried out at J.S. Hamilton Poland Test Laboratory, a certificate no. LT/606.3/2023 was issued confirming the compliance of the housing, made of 316L stainless steel of dimensions 246x95x41mm, with the IK06 and IP34 resistance and protection requirements.

*Fig. 1 Assembly drawing of the housing made of 316L.*

*Fig.* *2 Photo of the final housing construction with the electronics installed.*

Test setup and experimental results

Due to the potential purpose, a prototype environment in the form of a ventilation duct was built for testing the system and algorithms as shown in Fig. 3. It was used for research on the noise reduction system in ducts, as well as for research on acoustic phenomena, including destructive interference. Such a prototype allowed for conducting research in controlled conditions with an approximate known sound propagation path. The ventilation duct consisted of a pipe 200 cm long and 200 mm in cross-section, a Y-shaped tee, microphones and sound sources.

Two sound sources were installed in the ventilation pipe, and each of them fulfills a specific function. The speaker placed in the parallel axis of the channel simulates a noise sound source, while the speaker mounted on the tee acts as an ANC source. The speakers used in the prototype are characterized by a wide frequency response (38Hz-18kHz). In addition, they have a diameter of 200 mm, which matches the dimensions of the pipe so as to best cover its cross-section and create acoustic coupling between the membrane and the channel. The speakers were also equipped with housings that fulfilled a dual purpose.

*Fig. 3 Prototype diagram of the ventilation duct, with noise source, ANC speaker, reference and error microphone.*

*Fig. 4 Comparison of active noise cancellation system efficiency for in-channel systems without active ANC (blue line) and with active ANC (red line).*

The developed solution for duct system applications was tested. In order to optimize the system operation, tests were conducted in various configurations. Fig. 4 shows the effectiveness of the system without (blue line) and with active noise control system (red line) in the amplitude-frequency characteristic in the range up to 1250 Hz. The measurement was performed at the duct outlet location in close vicinity to error sensor.

In the case of measurements recorded for the system without the ANC system. The measurements are characterized by a local maximum for the frequency of about 390 Hz and its subsequent harmonics of about 780 Hz and 1170 Hz. In between, the signal spectrum is characterized by peaks of lower amplitude. The graph shows a significant reduction in the sound pressure level for the measured signal after the ANC system was activated. In the case of the first dominant frequency, the reduction was about 13.2 dB. For the second dominant frequency the obtained reduction was about 8 dB.

IP protection

As part of the developed strategy, it was decided to file a patent application. Single file covers two embodiments. Both examples fit into the developed strategy of protected claims. Within the application, the examples were carefully described and documented through supporting technical diagrams. Patent application was submitted and according to the description presented in the application, this procedure was initiated in order to obtain legal protection for the presented innovation. The patent application was sent to the Patent Office of the Republic of Poland and registered under P.447001 number titled “Active noise reduction system based on machine learning algorithms for industrial applications”.

Acoustic monitoring

Essence of the Solution

The Silencions Active Noise System is perfect solution for narrow band noise problems in low frequency. Additive manufacturing is used in Silencions to produce electronics housings and ANC components. Developed solution can be used to efficiently reduce noise of the source system based on developed electronic system and dedicated software. Developed system was tested in laboratory conditions and could be exploded for in real-world applications.

WANT TO KNOW MORE?

If you have technical questions or want to know more, let’s talk!

We have been monitoring, diagnosing and solving noise and vibration problems for years. We create acoustic adaptations for special tasks.