[0:00] hey everyone for my next project i need
[0:01] to get audio data into the esp32
[0:04] to do this we’re going to use the
[0:06] built-in analog-to-digital converters
[0:08] on the esp32 now the esp32
[0:12] integrates two 12-bit analog to digital
[0:14] converters
[0:15] adc1 has 8 channels and adc 2
[0:18] has 10 channels adc2 is used by the wifi
[0:22] driver so we can only use
[0:23] adc2 when not using wi-fi in addition
[0:26] some of the pins attached to adc2 are
[0:28] used for strapping pins
[0:30] so it cannot be used freely i’m going to
[0:31] be using wifi in my project so i’ll
[0:33] stick to using
[0:34] adc1 for simple low frequency sampling
[0:37] we can use the arduino analog
[0:39] read or we can use the expressive adc
[0:42] functions directly if you need
[0:44] very accurate readings then you can
[0:45] calibrate your analog to digital
[0:47] converter the chips manufactured
[0:49] recently this may have been done
[0:50] in the factory but if required you can
[0:52] also do this manually check the
[0:54] description
[0:55] for some links on how to do this to get
[0:56] a calibrated value from the adc we need
[0:59] to do a few steps we need a full back
[1:01] value for the vref
[1:02] if nothing is available from the
[1:03] calibration system we then need to set
[1:05] up our adc
[1:06] in a known configuration in this example
[1:09] code we are setting our adc
[1:10] to 12 bit resolution giving us a range
[1:13] from naught
[1:13] to 4096 we’re also setting our
[1:16] attenuation
[1:17] to 11 db this should give us the full
[1:19] input range
[1:20] from naught to 3.3 volts we can now pull
[1:23] back
[1:23] the calibration characteristics for
[1:25] these settings and now
[1:26] we can get the raw value from the adc
[1:29] and map it onto a voltage
[1:31] using the calibration characteristics to
[1:33] test this
[1:34] i’ve hooked up a potentiometer to one of
[1:37] the adc
[1:38] channels and printed out both the raw
[1:41] value
[1:42] and the calibrated value in a loop
[1:46] as we vary the voltage on the pin the
[1:49] value read from the adc
[1:52] changes
[1:58] you can see from my dev board that it’s
[2:00] been factory calibrated
[2:02] with a vref value using the adc in this
[2:07] way is fine
[2:08] if you only need to get a value every so
[2:10] often or you want to sample
[2:12] at very low frequencies for very low
[2:16] quality speech
[2:17] we need to capture a bandwidth of around
[2:19] four kilohertz
[2:21] for good quality speech we need around
[2:24] eight kilohertz
[2:25] and for high quality audio we need a
[2:27] bandwidth of 20 kilohertz
[2:31] to avoid issues with aliasing we need to
[2:34] sample at the nyquist rate
[2:36] which is twice the highest frequency you
[2:38] want to sample
[2:39] for audio this would be around 40
[2:41] kilohertz
[2:42] you can see in this animation the effect
[2:45] when the sampling rate
[2:46] is too low as the input frequency
[2:49] approaches past the nyquist limit we
[2:52] start to see aliasing
[2:53] and the output signal cannot be
[2:55] reconstructed from the samples
[2:59] since our audio signals will go up to 20
[3:01] kilohertz
[3:02] we need to sample at a minimum of 40
[3:05] kilohertz
[3:07] now doing this in a loop on the cpu
[3:09] would leave us no time
[3:11] for doing any other work our cpu
[3:14] would be constantly polling the adc for
[3:16] new samples
[3:17] and would not have much time for any
[3:19] other processing
[3:21] we would also struggle to achieve a
[3:22] consistent sampling period
[3:24] leading to weird artifacts in our audio
[3:27] fortunately
[3:28] there is an alternative mechanism for
[3:30] reading samples from the adc
[3:33] we can use the i2s peripheral to read
[3:35] the samples
[3:37] this has a dedicated dma controller that
[3:40] allows us to stream samples
[3:41] straight into ram buffers independently
[3:44] from the cpu
[3:46] i’ve written a simple sketch that uses
[3:48] i2s
[3:49] to read the samples and as audio data
[3:52] becomes available
[3:53] it streams it to a server running on my
[3:56] desktop computer
[3:57] which writes the audio data to disk
[4:01] i’ve set a sample rate of 40 kilohertz
[4:04] and allocated four dma buffers at 1024
[4:08] bytes each
[4:09] this should give us sufficient time to
[4:11] process any buffers without them being
[4:13] overwritten
[4:14] with new data we’ve set up the i2s
[4:17] peripheral
[4:18] to read from adc one channel seven this
[4:21] equates to
[4:23] gpio35 and then we set up a task
[4:26] to read the values from the its queue
[4:28] looking at our reader task
[4:31] it waits for an itos event rx done to be
[4:34] received
[4:35] and then reads bytes from the dma buffer
[4:37] into our own local buffer
[4:41] once we’ve read all the samples we’re
[4:42] interested in we do whatever processing
[4:45] we need to do
[4:46] and in our case i’m just sending the
[4:47] samples over to a local server
[4:50] running on my desktop machine here’s the
[4:52] output when we use our potentiometer
[4:54] to change the values on the adc for this
[4:57] example
[4:58] i’ve just used a sampling rate of 10
[5:00] kilohertz
[5:02] we can now move on to getting audio into
[5:04] the system
[5:05] i’ve got two microphone breakout boards
[5:08] the max 4466 and the max
[5:11] 9814 now the max 4466
[5:15] has an adjustable gain from 25 times
[5:18] to 125 times and the max
[5:22] 9814 has a built-in automatic gain
[5:25] control
[5:26] this will make quiet sounds louder and
[5:28] louder sounds
[5:29] quieter both these modules are simple
[5:32] and easy to wire up
[5:33] only requiring power and ground
[5:37] so here’s an audio recording from the
[5:39] max 4466
[5:47] testing testing
[5:55] as you can hear we have some audio data
[5:58] but it’s terribly noisy and it’s not
[6:00] going to be usable
[6:02] uh here’s the audio recording from the
[6:03] max 9814
[6:11] testing testing one two three
[6:17] it’s a bit better but there is still
[6:19] quite a lot of noise coming through
[6:21] it looks like the noise can inside with
[6:24] our transmission of data over wi-fi
[6:26] i’ve attached an oscilloscope to the 3.3
[6:29] volt output from the dev board
[6:31] and the output from the max 4466
[6:35] you can see that our 3.3 volt line has a
[6:38] lot of noise
[6:39] when the module is transmitting this
[6:42] noise is being amplified by the
[6:43] sensitive op amps
[6:44] on the microphone board looking at the v
[6:47] in line
[6:48] we can see a similar problem so what can
[6:50] we do
[6:51] we could connect up a separate power
[6:53] supply or we could use a battery
[6:56] to power our microphone boards but this
[6:58] is not going to be very convenient
[7:01] we could also turn off wi-fi but i need
[7:04] wi-fi
[7:04] for my project so the underlying problem
[7:08] seems to be the current draw when the
[7:10] esp needs to transmit
[7:12] this is causing a large amount of noise
[7:14] on the 3v3
[7:16] power rail which is then picked up by
[7:18] the very sensitive microphone amplifiers
[7:21] i’ve tried adding capacitors which have
[7:23] a small effect
[7:25] but to really fix it we would need to
[7:27] add excessively large capacitors
[7:30] to both the vin and the 3v3 rails
[7:33] so our microphone boards don’t require
[7:36] very much current
[7:37] so what we can do is take the vin line
[7:41] and pass it through an rc filter for my
[7:44] tests i’m just using a
[7:45] 30 ohm resistor and a 470
[7:48] microfarad capacitor we can then feed
[7:51] this filtered signal
[7:53] into a low dropout regulator to generate
[7:56] a clean
[7:57] 3v3 power supply for the microphone
[7:59] boards
[8:00] looking at the low-pass filtered vin we
[8:03] can see that we have a
[8:04] cleaner signal and on the 3v3
[8:07] output of the regulator we now have a
[8:10] pretty clean signal
[8:11] with noise down to around 20 millivolts
[8:14] instead of the 200 millivolts
[8:16] that we were seeing on the original 3v3
[8:18] line
[8:19] from the board looking at a trace from
[8:21] our microphone boards
[8:23] here’s the max 4466
[8:26] we don’t see any noise now when the
[8:28] esp32
[8:30] transmits and the same is true for the
[8:32] max
[8:33] 9814 so here’s an audio sample
[8:37] from the max4466 testing testing
[8:42] one two three and an audio sample
[8:45] from the max 9814
[8:48] testing testing one two three
[8:53] they are both better but still quite
[8:55] noisy
[8:57] one thing we can do is filter out some
[8:59] of the noise
[9:00] by over sampling and then take an
[9:03] average value
[9:04] as the actual sample value so i’ve tried
[9:06] using a
[9:07] median filter here and you can see
[9:10] that our simulation shows quite an
[9:12] impact
[9:14] and here’s the audio from the max 4466
[9:21] testing testing one two three
[9:24] and the audio from the max 9814
[9:28] testing testing one two three
[9:33] so which module should we actually use
[9:35] for my next project
[9:38] i think the max 9814
[9:41] comes out the winner it seems to be less
[9:44] susceptible to
[9:45] power supply noise and the automatic
[9:48] gain control makes it very useful
[9:51] we don’t have to fiddle with a
[9:53] potentiometer to change the volume
[9:56] um if you really need high quality
[9:59] low noise input then the built-in adcs
[10:03] on the esp32
[10:05] are probably not suitable and i would
[10:07] take a look at some external boards
[10:09] for this kind of project but on balance
[10:12] i think we’ll use the
[10:13] max 9814 for my next project
[10:17] so thanks for watching i hope you found
[10:19] this video
[10:20] useful and interesting all the code is
[10:23] in a github repo
[10:24] the links in the description if you did
[10:26] find the video useful
[10:28] then please subscribe to the channel and
[10:30] hit the like button
[10:32] there’s another video coming soon well
[10:34] i’ll actually do something with the
[10:36] audio data
[10:37] so see you in the next video