Interference Cancellation in Aircraft Cockpit by Adaptive Filters

This paper investigates on the development and implementation of adaptive noise cancellation (ANC) algorithm meant for mitigating the high level engine noise in the cockpit of an aircraft, which makes the speech signal unintelligible. Adaptive filters configured as interference canceller have the potential application in mitigat ing the above interference. A comparative study of Gradient based adaptive Infinite Impulse Response (IIR) algorithm and its modified version is performed using MATLAB sim ulator in terms of converging speed. From the simulation result the best IIR algorithm is used for implementation in Performance Optimized with Enhanced RISC PC (Power PC) 7448.


Introduction
High level of engine noise is present in the cockpit of an aircraft.Consider the commun ication of a pilot by radio fro m the cockpit to an air traffic controller, the engine noise would be interfered with the speech signal because the noise occupy the same frequency band as speech .The microphone into which pilot speaks also picks the engine noise, which makes the speech signal unintellig ible.A conventional filter cannot be used to remove the engine noise because the intensity and frequency of noise components change with engine speed, load and pilot's head.If a second microphone is placed at a convenient location in the cockp it, it is possible to obtain the amb ient noise field without pilot's speech.This noise could be filtered and removed fro m the transmission with the help of adaptive filters configured as noise canceller.There are some situations where a speech signal is interfered by periodic engine noise and no noise field free of pilot's speech is available.In such situations adaptive Line Enhancer (ALE) can be utilized as noise canceller.Th is paper focused on ALE as noise canceller An adaptive filter is a dig ital filter that can adjust its coefficients to give the best match to a g iven desired signal [1].There are different adaptive algorithms (eg.Least Mean Square (LMS), Normalized Least Mean Square (NLM S) etc) that can be used in time domain.Some widely used methods with Adaptive FIR filter have been explained by S. Haykins [2] & B.Wid row [3].Fir filters are mo re stable than IIR filters.Adaptive IIR filters are attractive: Many fewer coefficients may be needed to achieve the desired performance in so me applications [4].However, it is more difficult to develop stable IIR algorith ms, they can converge slowly.Adaptive IIR algorith ms are used in some applications (such as low frequency noise cancellation) where the need for IIR-type responses is great.In some cases, the exact algorith m used by a company is a tightly guarded trade secret.However a grad ient-based Adaptive IIR algorith m, with some additional features that enable it to adapt more quickly is exp lained by Don R.Hush [5].A modified version of the above algorithm g ives even faster convergence.So the mod ified algorith m can be used for implementation.

Adaptive signal processing
The term Adaptive can be understood by considering a system which is trying to adjust itself so as to respond to some phenomenon that is taking place in its surroundings.This is what adaptation means.Moreover there is a need to have a set of steps or certain procedure called algorith m by which this process of adaptation is carried out [6]. adaptive algorithm, which adjusts the variable filter to satisfy some predetermined ru les.In stochastic frame work (based on Wiener filter theory), the optimu m coefficients of a linear filter are obtained by minimizat ion of its mean-square erro r (MSE).A ll the adaptive algorith ms take the output error of the filter, correlate that with the samples of filter input in some way, and use the result in a recursive equation to adjust the filter coefficients iteratively.

Adaptive noise cancellation (ANC)
The ANC method uses a "primary" input containing the corrupted signal and a "reference" input containing noise correlated in some unknown way with the primary noise.
The reference input is adaptively filtered and subtracted from the primary input to obtain the signal estimate [1].p y p g

Figure 2. Adaptive noise cancelling concept
A signal s is transmitted over a channel to a sensor that also receives a noise n 0 uncorrelated with the signal.The combined signal and noise s+ n 0 form the primary input to the canceller.A second sensor receives a noise n 1 uncorrelated with the signal but correlated in some unknown way with the noise n 0 .This sensor provides the reference input to the canceller.The noise n 1 is filtered to produce an output y that is as close a replica as possible of n 0 .This output is subtracted fro m the primary input s+ n 0 to produce the system output z= s+ n 0 -y.Hence the output of the system contains the signal alone [1]- [7].

Adaptive line enhancer IIR structure
The Adaptive Line Enhancer (ALE) enhances the sinusoidal co mponent of the reference input so that output has high signal to noise ratio SNR.Fro m the ALE we also obtain an estimate of the sinusoidal frequency.This ALE can be utilizes as noise canceller.Several forms of the A LE are availab le.The most popular is th e FIR A LE.It has the advantage of being inherently stable and easy to adapt.However, it often requires a large number of filter weights to provide adequate enhancement of narrow-band signals.In an effort to reduce the number of weights, several forms of adaptive recursive filters (IIR) can be used.
The purpose of ' is to decorrelate the input noise and its delayed version present at the filter input.This causes the adaptive process to respond only to the sinusoid.As such, H (z) forms a band pass filter around the sinusoid, applying at the same time a phase shift to co mpensate for the delay so that the sinusoidal component of x k is removed at the summer.At the output of the adaptive filter x ' k,, the enhanced sinusoid is obtained.To simplify the analysis assume n k is white noise so that ' = 1 is sufficient for decorrelation.

Gradient based adaptive algorithm for IIR ALE
The technique used here is a gradient-based algorith m with so me addit ional features that enable it to adapt more quickly.The filter output is given by Where 0<< r <1 -2r<w<2r, r is the radius of the circle just inside the unit circle.Also from Figure 3, the error output is defined to be The coefficient update for the adaptive algorith m can be expressed as Where ρ is a parameter which controls the rate of convergence, ɛ k is the erro r, and α k is the partial derivative Fro m (1) and the definit ion of (3), recursive relationship for The coefficient update is then defined by (2),(3) and (5).The normalization factor is incorporated by reexpressing the coefficient update in (3) as Where k H and k D are as defined above and The "forgetting factor "v is in the range 0<<v<1 The performance of this a lgorith m can be enhanced even further by using an appro ximate partial derivative in place of α k in (5).The mod ified derivative, denoted by α mk ,is obtained fro m (5) by suppressing the recursive terms, i.e.,

Adaptive IIR noise cancellation
ALE can be utilized as noise canceller with experimental data.A LE is used as the basic structure for noise cancellation.Here the input signal X contains the speech signal and engine noise.The purpose of Δ is to decorrelate the input speech and its delayed version present at the filter input.This causes the adaptive process to respond only to the noise As such, H (z) forms a band pass filter around the noise, applying at the same time a phase shift to compensate for the delay so that the noise component of X is removed at the summer.By taking the autocorrelation of speech signal one can find out suitable value for Δ.So at the output of the summer (error signal) contains the speech alone.i.e., the engine noise is adaptively cancelled out.Let the signal to be interfered with noise such that the SNR is controllable and verify the performance of IIR structure in hig hly noise environment.The system is provided with the 05004-p.3permission to input a particu lar SNR (user's choice), a noise with corresponding amplitude is generated and added to signal.This noise corrupted signal will be processed.The Figure14 shows the comparison among the pure speech and noise filtered signals obtained fro m two algorith ms.Fro m the simulated result it is clear that even if the signal is interfered by a noise whose power is very high compared to signal power, adaptive IIR filter is able to pick out the noise.The blue co lored signal is obtained from normal algorith m.The red colored signal is produced by modified algorith m.Fro m the figure below it is clea r that the modified algorith m is converging faster than the normal one.Hence modified version of the gradient based algorithm is used for implementation on PowerPC.

Implementation
Interference cancellation using adaptive IIR algorithm is implemented in PowerPC 7448 SBC(Single Board Computer).

Implementation process flow
The input signal contains the speech signal interfered by engine noise.The purpose of delay is to decorrelate the input speech and its delayed version present at the filter input.This causes the adaptive process to respond only to the noise.So at the output of the summer (erro r signal) contains the speech alone.i.e., the engine noise is adaptively cancelled out.

Results and discussion
The interference cancellat ion using IIR filter has been implemented successfully on PowerPC 7448 with Modified Gradient Based Adaptive algorith m and verified the output.Here the signal applied to the input is speech corrupted by engine noise.After adaptive filtering process speech alone is produced at the output.

Conclusion and future scope
A detailed survey on ANC is carried out and observed that many fewer coefficients may be needed to achieve the desired performance fo r IIR filter when compared to FIR.Fro m simu lation under different noise environment, it is concluded that the modified gradient based algorith m is more suitable fo r Adaptive IIR noise cancellation.Even if the signal is interfered by a noise whose power is very high co mpared to signal power, adaptive IIR filter is able to pick out the noise.i.e., even in the highly noise environment adaptive IIR filter is powerful.
Future work includes the implementation of the pipelined version of the filter for improved speed, and to optimize the programmab le processing element at circuit level for efficient ASIC imp lementation of the reconfigurable fabric.

Figure 1 .
Figure 1.Basic adaptive filter structureThe adaptive filter structure shown in the Figure1in which the filter's output y is compared with a desired signal d to produce an error signal e, which is fed back to the adaptive filter.The error signal is given as input to the

Figure 3 .
Figure 3. Adaptive line enhancer [2] After performing the adaptive process, at the output of the filter we get the enhanced version of the sinusoid.At the output of the summer, (error signal) sinusoidal component of input is removed.The simu lation result shows the Power Spectral Density (PSD) of pure signal(Figure4), noise corrupted signal(Figure5), comparisons of PSD of error signals (Figure6)and enhanced signals(Figure7) using both the algorithms.Mean Square Error (M SE) of t wo algorith ms is also shown (Fig ure8).Fro m the PSD of error signal it is clear that a very large reduction of power occur at the output of the summer.The modified algorith m provides more reduction in power.

Figure 4 .
Figure 4. PSD of pure signal