A Pre-Filter for Hands-Free Car Phone Noise Reduction: Suppression of Harmonic Engine Noise Components

Abstract

Single channel noise reduction is a severe challenge, especially when considering the difficulties of estimating the appropriate spectral distribution of interfering noise. Usually, all approaches, e.g. Wiener filters, Ephraim-Malah filters, and Kalman filters, rely on the assumption that the interfering noise components are longer-term stationary when compared to desired signals. Here, we focus on the application of noise reduction for car phones meaning that the interfering noise is car noise with its specific properties. When analyzing car noise, one observes that the assumption of long-term stationarity is true only as a first approximation. Analyzing car noise in more detail, described later in this chapter in the second section, one discovers that the different components vary in dependence of the car velocity or the engine speed. Engine noise, in particular, exhibits clear and predicable properties. One can show that engine noise consists mainly of a harmonic structure with strong spectral components at multiples of one half of the engine frequency. This allows the design of methods for specifically reducing these engine noise components since the engine frequency is available in all modern cars on the internal CAN bus. The description and the analysis of different methods for specific engine noise reduction are the topic of this chapter. The target is to design a kind of pre-filter for the classic noise reduction approaches mentioned above and specifically pre-filter the engine noise components. This pre-filter facilitates the work of the classic noise reduction methods by removing the highly non-stationary and strong engine noise components. In this chapter we present two different types of pre-filter approaches for the reduction of the engine noise components: