*3.3. Acoustic Environment of the Nodes of the WASN*

There are currently 24 low-cost high-capacity sensors deployed in a WASN in district nine in Milan, as described in Table 1. Nevertheless, in the framework of the DYNAMAP project, the acoustic environment of the sensors has been studied [45] and, together with the description of several roads conducted by the colleagues of Universitá degli Studi di Milano Bicocca [46], we reached the conclusion that not all the acoustic data collected in all the sensors was suitable to conduct this kind of analysis. Under ideal conditions, the noise events under test should be recorded in an free-field environment; an unfeasible requirement, not only for the DYNAMAP project, but also for any other project that intends to collect urban noise events.

An analysis was conducted with the aim of determining which sensors could capture noise events in a similar way to this ideal free-field environment. The first item to consider was the placement of sensors in narrow streets with buildings, due to the acoustic effect over noise events, which substantially changes the frequency distribution of the recorded data. We finally discovered that two of the sensors (hb137 and hb145) were placed in a facade, but surrounded by parks. Sensor hb137 was located in a quiet area, surrounded by two parks with trees. Furthermore, the closest building to this sensor was more than 55 meters away (see Figure 4). Sensor hb145 was also in a narrow street, but was surrounded by a park; both of them were placed in public buildings, one of them being a school. The closest wall to the sensor was around 35 m away (see Figure 5). Nevertheless, in both locations, the sensors were placed on the façade, so the *façade effect* will appear for both recordings; in future work, the authors will consider working on its mitigation [50].

(**a**) Location of sensor hb137 (yellow square) (**b**) Sensor placement in the facade

**Figure 4.** Sensor hb137 location and picture.

(**a**) Location of sensor hb137 (blue square) (**b**) Sensor placement in the facade

**Figure 5.** Sensor hb145 location and picture.

Once the sensors had been chosen, an analysis of the recorded anomalous noise events was conducted. Dozens of types of anomalous noise events were found to be recorded by those two sensors, such as: Birds (609 occurrences), brake (62), truck (20), siren (15), door (131), people (169), trolley (10), dog (134), aeroplane (28), chain (21), step (45), bell (11), bike (16), glass (3), bus door (8), saw (8), and tram (16). In Figure 6, four examples can be found of anomalous noise events recorded in sensor hb137 and hb145, showing that different ANEs present very variate frequency distributions.

The analysis of the available samples of anomalous noise events was focused on the spectral diversity and minimum length of the events. The events used for this analysis, including those from both sensors (hb137 and hb145), had an overall length of 4865 s. The maximum length of the processed ANEs was 191 s and the minimum length was 0.1 s. Another of the pursued goals was to study several types of spectral distribution, both wide-band and narrow-band.

In Figure 6, the bird spectrogram in Figure 6a presents a short audio of 0.5 s and a distribution of frequencies between 5 kHz and 7.5 kHz. The siren spectrogram in Figure 6b shows a good signal-to-noise (SNR) ratio in the raw audio, which clearly presents a Doppler effect. Even more, it was a long event, with a duration of around 25 s. In Figure 6c, a good signal-to-noise ratio brake is plotted, with a clear component around 11,000 Hz. This event was around 2.5 sec length. In Figure 6d, we can observe a wide-band signal, whose spectrum includes frequencies up to around 20 kHz. Low frequencies, between 0 Hz and 1 kHz, are occupied in all the audio pieces by road traffic noise, which presents energy mainly in those frequencies.

**Figure 6.** Examples of anomalous noise events recorded by sensors hb137 and hb145.

### **4. Channel Model Design**

In this section, we define three urban channel models, computed as a linear combination of the input source and attenuated delayed paths. The three channel models are static (i.e., neither the transmitter nor the receiver changed their position while the sound signal was propagating) and, hence, no signal frequency spread and Doppler shift are expected. We defined three channel models which represent three different urban scenarios, with different number of paths and delay. The aim is to transmit the sound signals through each of the channel models and study the changes of time and frequency response.

Prior to applying the recorded sound signals to the three channel models, we characterised the channel in terms of time-spread by means of a PN sequence with good auto-correlation and cross-correlation properties [37]. The goal was to be able to detect each of the paths of the channel impulse response with a wide enough bandwidth to check whether the input sound signal had suffered any impairment throughout its whole bandwidth.
