Lab 2

Before we are able to determine the presence of a 660 Hz signal in the robots environment, we need to wire the microphone sensor correctly and start receiving digital pulses encoding the audio impulses around the circuit. We made a simple inverting amplifier with ~100x gain based on Team Alpha's design. A schematic of it is shown below. We originally tried a non-inverting configuration, but the output's DC bias kept slowly creeping upwards, making the amplified output clip.

The microphone takes an audio signal and it is sent through a DC blocking capacitor. Then, this signal is amplified with an inverting amplifier that has a DC bias at 0.5*V_cc. Since V_cc was 5V for us, the op amp output is centered at 2.5 V (so, halfway between the power rails). This allows for the max signal swing at the output without clipping.

We combined a simple RC low pass filter after the output of the amplifier to block out background chatter and any decoy frequencies. When putting the amplified microphone signal on the scope, we saw a spurious ~1300 kHz frequency so we made a low pass with a R = 12 kΩ; and C = 100 uF to get a cutoff frequency of \( 1 \over 2 \pi RC \)= 1326 Hz. A picture of the microphone signal amplified is shown, with the microphone signal in blue and the amplified signal in yellow. FFTs of the whole system is also shown, unfiltered on the top and filtered on the bottom. From this, it can be seen that the filter attenuates the spurious frequency a bit.

Finally, we were able to use the Arduino FFT Library to begin detecting the 660 Hz signal that we were searching for. We find the max point of FFT magnitudes at 660 Hz. From this point, all we had left to do is the creation of an automatic detection algorithm for both audio and IR signals.

When running the FFT, we saw that the 660 Hz signal was in bin 6. This led us to believe that the Arduino was actually sampling ~67 kHz instead, because for 256 samples, the 255th sample = \( fs \over 2 \). So, each bin is \( fs \over 2\)\(1 \over 256\). Since the 660 Hz was in bin 6, 660 (Hz) = \(fs \over 2\)\(1 \over 256\)\(6\) (Hz). Note the the FFT shown was not normalized.

We were able to begin sensing IR light with a very simple setup. We place our Vout node connected to a pulldown resistor, separated from Vdd by the phototransistor sensor. This design rests at 0V in the absence of IR light, and rises as sensed IR light allows current to flow through the circuit and a growing voltage drop is measured across the resistor. Increasing the resistance here would make circuit more sensitive, but for our needs a 2kΩ was enough.

The rationale behind this decision was relatively straightforward. The signal we were picking up off our IR sensor would fall to 0V i n the absence of light, but rests very close to 0V when detected IR light isn't very strong as well. Amplifying our sensor output would therefore allow us to see the consistent 6.08 kHz signal we are looking for. However, the 18 kHz signal's proximity in the frequency domain to 6.08 kHz would result in some saturation of the signal we're searching for. This is because there were a lot of harmonics from both signal sources. In order to avoid this issue, we implemented a low pass filter with a pole of ~16 kHz to reduce the amplitude of the decoy signals and noise our robot would have to parse through. (R = 10 kOhm, C = 1 nF so f_c = 15.9 kHz). Pictured above you can view our FFT analysis of a 6.08kHz signal before amplification and filtering. Below you can view the post amplification and filtering results.

Determining the presence of a 6 kHz signal right now is done using the FFT library. We find that the signal 6.08 kHz signal appears in bins 44. The decoy signal appears in bin 122. This roughlyagrees with what we saw earlier, with a f_s ~67 kHz because f_{detect_robot} = (f_s/2)*(1/L)*(bin_number), where L = number of samples (256) and same for f_{decoy}. So, f_{detect_robot} = 5.808 kHz and f_{decoy} = 16kHz, which is off but might have been the next highest peak detected due to the harmonics of both the robot and decoy robot falling into the same frequency bin and adding up. Because we generally see a large amount of environmental IR light, we only say that a 6.08/18 kHz signal is present when the perceived amplitude is of a predefined threshold in magnitude. This threshold is determined by analyzing all FFT frequencies and observing the largest peaks, then calculating a relative threshold based off these values. The program logic then assesses if the bins where we expect to see the whistle signal (660Hz), IR hat (6.08 kHz), and IR Decoy (18 kHz) contain FFT magnitudes that surpass the threshold we defined.