The typical noise levels inside a fighter aircraft cockpit will be broadband and will be in the range from 95 to 105 dB. This level of noise exceeds the OSHA requirement of 85 dBA, 8 h/day exposure criterion, which may affect the alertness of the pilot and prolonged exposure may result in hearing impairment. Increased rate of periodontal diseases and cardiovascular risks in aircrew members because of high cockpit noise demands the reduction of noise levels [1–4]. Further, cockpit noise reaching the pilot’s ear through radio communication channel affects the intelligibility of speech and makes the identification of caution signals difficult [5,6].

The primary objective of a pilot helmet is to provide safety. But, it also provides protection against certain amount of high frequency noise owing to passive noise attenuation achieved through its design and fitment [7]. However, considerable amount of low frequency noise will still penetrate the helmet and this can only be addressed with Active Noise Control (ANC) technique.

A few ANC headsets discussed in [8–16] perform satisfactorily in aircraft cabin environments and helicopters. But to address noise inside a fighter aircraft helmet, the ANC algorithm needs to be more specific and dynamic to handle high level of noise entering through multiple paths. The noise appearing in the helmet earcups is because of two components, one is the low frequency noise that enters through the helmet directly, referred to as Cockpit Noise Direct (CND) and the other one is the noise that enters through the pilot microphone via Audio Management Unit (AMU), denoted as Cockpit Noise through AMU (CNA). Currently existing Helmet ANC systems for fighter aircrafts focus only on CND [17–19]. In this paper CNA is also addressed along with CND.

Although tremendous efforts have been made by the scientific community to develop a better ANC system using machine learning algorithms like genetic algorithm, fuzzy logic, artificial neural network, etc. [20–25], the traditional adaptive algorithms have always proven to be more stable, computationally less intensive and quite attractive to be used in real-time applications. Generally, noises like CND is addressed using FxLMS based Feed Forward ANC (FFANC) algorithm [26]. But here, as the noise is broadband in nature, better tracking of ANC system is necessary with respect to the varying noise levels and characteristics of the aircraft noise. For this, various adaptive ANC algorithms are proposed [27–32]. In this paper, a Variable Step Size Griffiths (VSSG) algorithm for FxLMS algorithm as described in [33] has been used to achieve a faster convergence with lower residual noise. VSSG algorithms and its variants [34,35] have been proven for improved performance of LMS and FxLMS algorithms [36,37]. The second noise component, CNA is mitigated using an LMS based adaptive noise canceller.

In a typical ANC system, ANC algorithm starts execution once the system is powered ON, irrespective of the presence/absence of noise. This leads to unnecessary power consumption affecting the battery life. To tackle this, reference signal energy is used in this paper for turning ANC ON. Additionally, one of the major failure condition in the ANC system is the occurrence of divergence, which leads to the pilot being exposed to a higher level noise and deteriorated communication signals. This is a major safety issue and has to be addressed effectively. Signal energy [38,39] based tracking algorithm is proposed in this paper to address this issue. The algorithms proposed are targeted to meet the real-time requirements (majorly, less computational complexity and lower power consumption) of an embedded application. Further, the implementation of the algorithms is carried out on a Texas Instruments TMS320C6748 floating point DSP processor and its performance is demonstrated in a helmet ANC experimental setup.

Fig. 1 shows the block diagram schematic of helmet ANC system. As discussed, the two noise components, CND and CNA are addressed using FFANC and adaptive noise canceller respectively. The cockpit noise picked up by reference microphone is used as noise reference x(n) for both FFANC and adaptive noise canceller. FFANC is based on VSSG FxLMS algorithm and has two separate channels for each earcup. The generated antinoise is fed through the audio communication speakers present inside the earcups. Adaptive noise canceller is based on LMS and its output (filtered pilot microphone signal) is fed to the same speakers. The residual noise is captured using error microphones placed inside the earcups. The algorithms to facilitate the dynamic operation of ANC so as to handle the scenarios in aircraft noise are depicted in Fig. 2 and are discussed in following subsections.

For simulation of ANC algorithms, the cockpit noise of a Light Combat Aircraft (LCA) was recorded using M-Audio digital noise recorder. The recorded data consist of temporal noise during the takeoff, flight and landing conditions. The noise levels were recorded using a B&K 2270 sound level meter, which showed that the noise levels vary between 90 dBA−110 dBA . From the periodogram computed using MATLAB and one third octave spectrum computed using DATS analysis tool indicated that the noise spectrum was of broadband nature with significant contribution from low frequency components.

The algorithms discussed in Section 2 are simulated in MATLAB with parameters given in Table 1. A practical impulse response of length 128 samples is used for VSSG-FXLMS and adaptive noise canceller. The comparison of ANC with and without VSSG can be seen in Fig. 5.

In Table 1, μ is the step size for ANC without VSSG. In ANC with VSSG, initially the step size is μmax . Towards convergence, the step size keeps on decreasing and is stagnated to μmin to stop further reduction. In Fig. 5, improved performance of ANC using VSSG-FXLMS can be observed with respect to convergence time and residual error.

A speech signal corrupted by fighter aircraft noise is considered for simulation of adaptive noise canceller. Fig. 6 shows the convergence of adaptive noise canceller.

Fig. 7 shows the simulation of algorithms for battery saving and system failure management, which are included to control ANC algorithm. It can bee observed that whenever ANC diverges, the main path filter coefficients are reset and antinoise starts to build from initial conditions. It can be seen that as the power of the reference signal varies, ANC system stabilizes itself on its own.

For effective noise cancellation, antinoise must be available by the time the noise signal arrives at the speaker. It implies that the acoustical delay must be greater than the electrical delay. This is referred to as causality condition. This is a time constraint imposed by the ANC system and the real-time implementation of ANC algorithms must adhere to this condition. Fig. 8 depicts the delays involved in both acoustic and electrical domain.

The acoustic delay from the input microphone to the speaker is given as,(18) A=L C0 where,

A–Acoustical delay.

L–Distance between reference microphone and speaker in meters.

C0 –Speed of sound in air in m/s.

Electrical delay is due to the delay involved in antinoise generation. The delay path involves the electronics and DSP processor. Apart from the application loop time of DSP, the lowpass filters contribute significantly to the electrical delay.

In case of helmet ANC, for typical distance of 20 cm between reference microphones to speaker, the causality condition states that the DSP should operate at 14 kHz, for 8^th order butterworth filter [26]. To facilitate this condition, the algorithms have to be coded using processor specific assembly language only.

As shown in Fig. 1, the reference noise from the cockpit is picked up using a high performance electret microphone. Microphones are fitted inside the helmet earcups for fetching error signals. The communication speakers fitted inside the earcups are used for ANC system as well. The ANC unit consists of a low power high performance DSP board based on TMS320C6748 and a signal conditioner board which consists of pre amplifiers for microphones, low pass filters, A/D and D/A converters, anti-aliasing filters, reconstruction filters and power amplifier for loud speakers. It also has an SD card slot for recording the residual noise inside the helmet earcup (error microphone signal) for offline analysis of ANC performance. The ANC unit is powered by a 9 V rechargable battery. The AMU simulator mixes the pilot microphone signal with the communication signals received from Air Traffic Control (ATC).

To meet the computational requirements of the algorithms, they are coded with pre-estimated secondary path coefficients on DSP processor using processor specific hand optimized assembly language coding. With optimized coding, s sampling frequency of 14 kHz was achieved which greatly increases the processing speed of the DSP. The operating noise frequency range of ANC is largely affected by the the execution speed of ANC algorithm on the DSP board. As the Nyquist sampling criteria suggests: to avoid aliasing, the operation bandwidth should always be less than half of the sampling frequency, i.e., B<fs/2 [40]. Hence the ANC cutoff frequency was set to 4 kHz (i.e., ANC will satisfactorily operate to the noise components having frequencies upto 4 kHz).

The experiments were conducted in an acoustically treated semi-anechoic chamber at CSIR-NAL (Fig. 9). Recorded cockpit noise is played using a 100W loudspeaker and the audio signals from error microphones are recorded in the SD card during ANC is Off and On.

The Power Spectral Density (PSD) of the cockpit reference noise, residual noise signals during ANC is Off and On are computed. The Noise attenuation in terms of decibels at particular frequencies is shown in Fig. 10. It can be observed from Fig. 10 that the helmet filters out much of the high frequency noise components of cockpit noise (passive noise attenuation) and the ANC unit effectively handles low frequency noise components within 4 kHz.

The ANC system performance is normally defined as the attenuation achieved in terms of Signal to Noise Ratio (SNR) in dB, which is computed using,(19) SNR=10 log10  ∑n=1N⁡eeANC On(n)2/∑n=1N⁡eeANC Off(n)2

The passive and active noise attenuation achieved by the Helmet and dynamic ANC respectively are tabulated in Table 2.

The ANC helmets and headsets discussed in [8–16], give passive noise attenuation of 10–15 dB and active noise reduction of 5–10 dB. Compared to these, the proposed dynamic ANC system gives passive noise reduction of 26 dB and active noise attenuation of 18.06 dB as seen from Table 2.

In this paper, a dynamic ANC system has been presented which proves suitable for a fighter aircraft pilot helmet. The ANC unit caters the noise entering through helmet using FFANC. The noise entering through AMU is catered using adaptive noise canceller. The robustness of the system was increased by using VSSG FxLMS algorithm which facilitates faster convergence and lower residual noise. Energy based detectors used for ANC operation control saves battery charge life cycle. The processor specific assembly language used for coding algorithms facilitated higher sampling rate enabling the generation of more samples of antinoise per second. The real-time evaluation shows that the dynamic helmet ANC system is capable of attenuating significant amount fighter aircraft cockpit noise.