Internship project: Development of Feedback Suppressor (FBS) for hearing aids

Start: from May 2017
Duration: 3 to 6 months
Location: Videolab, Strijp-S Torenallee 20
Contact: Ad van de Voort, Videtur BV Eindhoven
E: ad.van.de.voort@videtur.nl
T: 06-25077205

The internship is inside a startup that introduces a new product: a low entry hearing aid
with a high quality sound. The smart phone is used as a high performance computing
device so that it can run the state-of-the-art DSP algorithms in order to bring the most
optimum sound to the user.

Because of their high gain hearing aids suffer from feedback between earphone and
microphone that causes instability and this instability manifests itself as a whistling
sound. The subject of this internship is to develop an algorithm that suppresses the
feedback: the Feedback Suppressor (FBS) algorithm for the smart phone hearing aid.

The FBS algorithm has the following requirements:
- the Added Stable Gain (ASG) of the FBS must be 10 dB, but preferably more (ASG is the possible increase in gain before instability when compared to a system without FBS)
- the FBS must have a fast convergence: ASG of 10 dB after at most 500 ms
- the FBS must provide a real-time indication of its performance by providing the realized ASG at any moment so that the volume of the hearing aid can be turned down when the specified ASG is not yet reached
- during operation no extra signals may be added

The FBS has additional requirements due to the system/hardware on which the FBS runs:
- the FBS must be able to deal with a non-linear forward path: the forward path is a multi-band compressor
- the FBS must be able to deal with high latency (about 30 ms) and an unknown exact latency in each run (the precision is about 2 ms)
- the sample frequency is 48 kHz or 44.1 kHz, it is possible to apply sample rate conversion in order to have a lower sample frequency, in that case the sample frequency must be at least 16 kHz
- the FBS must be able to deal with a changing feedback path due to the changing environment, the feedback path of the smart phone hearing aid can be measured with a special app
- the FBS uses double precision floating point calculations
- the computational complexity must be limited (as a very rough indication: at a sample frequency of 48 kHz a Block Normalized Least Mean Square, BNLMS, algorithm with block size 32 and weight vector size 64 should be possible)

Activities:
- design the FBS algorithm
- simulate and test the FBS on a numerical analysis tool (for instance Scilab)
- port the FBS algorithm to the C-language and perform simulations, tests and unit tests on a numerical analysis tool
- test the FBS algorithm on the smart phone

What can you expect:
- develop, apply and test the latest DSP technologies
- work in a professional environment with a high degree of freedom and independence
- work close with the director and witness all aspects of a new product introduction
- work in an inspiring environment
- a chance to develop algorithms that will be used in a consumer product