|
Hi All,
?
Do you know of any designs or indeed PCB modules for active noise cancelling in a microphone?
?
I repeatedly encounter two situations recently where background acoustic noise is a significant problem when needing to use a microphone. Using a dynamic mic helps a bit, but still fails to cull enough noise. It has gotten me curious about whether there's an active noise cancelling solution.
?
A) Commentating on a rowing competition:
This involves cycling down a towpath, with a headset mic on, meaning weight and size limitations. Spectators are typically standing alongside the towpath, cheering. Bank party crews use whistles and air horns to signal to the rowers distances. So all of this noise is largely within the speech frequency band and is coming from in front, behind and to one side of the cycle commentator.
?
B) Flying a General Aviation aircraft:
To keep aircraft weight down most light aircraft don't have much acoustic dampening between the engine and the cockpit. So the cockpit is rather noisy both acoustically (and electromagnetically). To communicate with each other and air traffic control via the radios, we use headsets with microphones. The headphone side of things is usually passively noise cancelled at minimum, and many people opt for actively noise cancelled headphones. However by the looks of things, the mic side of things is less clever: it's typically a cardioid pressure gradient electret, or a cardioid dynamic. The noise is predominantly coming from the front, with a particular prominence in the lower frequency end of the speech band from the engine. It's not lost on me that most unidirectional mics tend not to be that directional in the lower frequency range.
(NB: there are no certification requirements for the mic/headphone side of the comms equipment for the aircraft I fly)
?
It got me wondering:?
Would some sort of active noise cancelling for the mic be possible, or is a single directional mic transducer plus a speech band-pass filter the best we can hope for?
?
Is there a good speech processing DSP chip which can identify a speech signal, and chase the signal frequencies, filtering off other frequencies??
?
Could we use multiple small transducers and employ beamforming? If so, are there any PCB modules doing this (e.g. from conference room boundary mics)
?
In other words, how would you achieve real-time speech isolation in these noisy environments?
?
|
It's been known for decades that one of the best noise-cancelling mic arrangement for voice pickup is two identical mics. One (A) is positioned right at the mouth, the other (B) is a few inches away. When you subtract the signal from B from the signal from A, what's left is mostly voice, since A was much closer to the mouth. I believe that many aircraft headsets do this for noise canceling at the mic.
?
In an older mic book (Lou Burroughs, "Microphones; Design and Application"), the author shows two omni dynamic stick mics (eg EV 635A) reversed then taped together, with one mic head beside the other's XLR end. A y-cord is made that connects the two mics out of phase. The user - typically a reporter in a noisy location - talks closely into one of the mics.
|
There are several noise-cancelling capsules from JLI.
They are based on a delayed path for the backwave.
Of course, they need to be very close to the source.
toggle quoted message
Show quoted text
Hi All,
?
Do you know of any designs or indeed PCB modules for active
noise cancelling in a microphone?
?
I repeatedly encounter two situations recently where
background acoustic noise is a significant problem when needing
to use a microphone. Using a dynamic mic helps a bit, but still
fails to cull enough noise. It has gotten me curious about
whether there's an active noise cancelling solution.
?
A) Commentating on a rowing competition:
This involves cycling down a towpath, with a headset mic on,
meaning weight and size limitations. Spectators are typically
standing alongside the towpath, cheering. Bank party crews use
whistles and air horns to signal to the rowers distances. So all
of this noise is largely within the speech frequency band and is
coming from in front, behind and to one side of the cycle
commentator.
?
B) Flying a General Aviation aircraft:
To keep aircraft weight down most light aircraft don't have
much acoustic dampening between the engine and the cockpit. So
the cockpit is rather noisy both acoustically (and
electromagnetically). To communicate with each other and air
traffic control via the radios, we use headsets with
microphones. The headphone side of things is usually passively
noise cancelled at minimum, and many people opt for actively
noise cancelled headphones. However by the looks of things, the
mic side of things is less clever: it's typically a cardioid
pressure gradient electret, or a cardioid dynamic. The noise is
predominantly coming from the front, with a particular
prominence in the lower frequency end of the speech band from
the engine. It's not lost on me that most unidirectional mics
tend not to be that directional in the lower frequency range.
(NB: there are no certification requirements for the
mic/headphone side of the comms equipment for the aircraft I
fly)
?
It got me wondering:?
Would some sort of active noise cancelling for the mic be
possible, or is a single directional mic transducer plus a
speech band-pass filter the best we can hope for?
?
Is there a good speech processing DSP chip which can identify
a speech signal, and chase the signal frequencies, filtering off
other frequencies??
?
Could we use multiple small transducers and employ
beamforming? If so, are there any PCB modules doing this (e.g.
from conference room boundary mics)
?
In other words, how would you achieve real-time speech
isolation in these noisy environments?
?
|
Here's one circuit that would employ two electret capsules in an noise-cancelling arrangement. We would of course use a better op-amp, and maybe adapt the OPA-Alice circuit to use phantom powering here.
?
|
RE: Noise Canceling Microphones/headsets
____________________________________________________
I concur with Ken J.
What you seek has been worked over pretty well, especially by some commercial outfits, Andrea-Electronics comes to mind, as does Knowles, Inc., and maybe be VXI, Inc. (the latter is famous for wireless BT headsets favored by long haul truckers) and they have all made active and passive noise reducing microphones for computer speech recognition applications, e.g., DragonSystems/Nuance Dragon NaturallySpeaking products.)
Active sets typically use equivalent mic capsules pointing in opposite directions, so sounds hitting both capsules equally tend to be canceled (because they are out of phase) while intended speech is captured unequally by the two capsules, so while there may be some out of phase cancellation, the sound is mostly conveyed to the target application because it is not fully canceled.
I contrast, what many manufacturers call "passive noise cancellation" is just close-talking a unidirectional microphone. You know, the good old cardiod pattern used close up tends to pick up what you say, and "reject" (i.e., not hear as loudly) ambient sound in the larger venue.
Lots of work has been done in this field for the Air Force and other noisy situations, including beam forming for automobiles - see especially Andrea-Electronics which has, in the past, published multiple White Papers on the topic, as it is a sub-contractor to the automotive and aerospace industries.
Therefore, Google (or a better search engine) is your best research assistant, as there is a lot of research on this topic.
How do I know? Because I have lunch every month with a guy who performed speech intelligibility research and helped develop speech recognition systems for the Air Force, and because I was a product field tester for several speech recognition equipment vendors, manufacturers, and software companies using same, including, but not limited to, various online retail mic vendors, Telex, DragonSystems, IBM, Microsoft, Lernhout&Hauspie, Kurzeweil AI, and so several others.
Be cautious and wary of manufacturer and vendor claims about noise cancellation - in my view, much is just exaggerated, overblown sales puffing. Be diligent and research - I believe the answers are mostly in the public domain, and not proprietary trade secrets.
Just my take. Your mileage may vary. James
|
These days it's done digitally using DSP, and it can be easily done in real time.
?
The principle is: you have two sound sources, one is a mic that is (mostly) capturing the speaker, like a directional mic close to the speaker's mouth (the primary sensor), and one mic is (mostly) capturing the noise (the reference sensor), e.g. an omni mic somewhere near the speaker (but preferrably not directly in the direction the speaker is talking in).
?
An adaptive filter algorithm then continuously adjusts filter coefficients that make the filtered reference signal (which is mostly the noise) match the primary signal as closely as possible. Because the reference signal is mostly noise, this filter will basically make the noise from the reference sensor look like the noise picked up by the primary sensor, compensating for frequency response and phase differences. But because the speech is mostly missing from the reference signal, when the filtered reference signal is subtracted from the primary signal you are mostly left with the speaker's voice.
?
See for a starting point.
|
Hi,
?
This is a pretty late reply but I'll share what little knowledge I have in the field of speech enhancement. I apologize in advance if the information posted here is technically wrong or misleading as I have spent only a short amount of time on a research project concerning this topic.
?
In the case of single channel and considering the scenarios described above:
?
There are a plethora of methods in the current research space but the core concept has always been to use statistical methods to find a transformation such that the input mixture (noisy signal) can be segregated into either the speech waveform, noise waveform or both. This transformation can be done through a variety of methods such as: Linear systems, spectrogram based neural networks, end-to-end time-domain synthesis (Probably the one with the largest interest at the moment due to the advances in GEN-AI and computing resources) and more. In the case of Low-SNR speech enhancement most of the methods outlined above perform much better at cancelling out stationary/ergodic processes such as those found in aircraft noise (mostly mechanical with low variance in observed noise over time) compared to babble noise as the probability density function of speech from multiple speakers tend to be the same.
?
Multi-channel:
?
Multi-channel methods are an extension of single channel methods with the introduction of other signal sources and necessitating phase alignment with reference to the position of the source signal (speaker's voice). This method allows for 'pre-enhancement' of the incoming mixture through the subtraction of the phase aligned input signals. The 'pre-enhanced' signal can then be forwarded to the developed transformation scheme to further enhance the noisy mixture. Research that has focused on the use of multi-channel methods were found to perform better with even lower SNR speech signals especially for babble noise. Alternatively the incoming signals can be forwarded to the enhancement scheme to also derive the enhanced speech signal.
?
A well-known approach to start from would be the Wiener filter due to its simplicity to implement and mature research based around it. This filter applies linear power coefficients to the observed mixture to enhance the signal without modifying the phase. For optimal enhancement the?optimal power spectral densities of both the noise and speech must be known and forwarded to the filter. Deriving these optimal coefficients can be done through the method of codebooks, linear systems and neural networks.
?
Both of the methods proposed above can be implemented in 'real-time' it just depends on how fast your processing hardware is to perform the enhancement once the speech signal is received.
?
Once again I must apologize if I have posted technically incorrect information here. I hope that the information contained in this reply is of some use.
?
|
I thought I'd provide an update to all of this...
?
I decided to focus on the General Aviation scenario because I do that more often. I wanted active noise cancelling headphones and microphone but was too cheap to pay the ?1200 for a Bose A20. I had looked at aviation clip on mics (Uflymike / nflightmic etc), but these all seem to be single unidirectional capsules, plus nobody had them in stock.
?
The aircraft I fly uses a Garmin GNC255. This has inputs for two microphones. It supplies about 10V bias voltage on the ring of a 3 pole jack, which also serves as the audio input. It can only supply up to about 10mA.
?
I first played around with a few PCB modules:
?
1) AN-93 digital signal processing dual microphone (from AliExpress)
- Uses two electret microphones (source and reference)
- Pros:
- Surprisingly good at cleaning up speech
- Cons:
- Only worked with the specific (omni) FET electrets that were provided, refused to work with unidirectional FET electrets, even matched pairs with similar sensitivity
- Kit is designed for conference call environments: The stock electrets were prone to clipping when speaking at close proximity.?
- Current draw ~12mA, possibly too high to run off bias supply
?
2) JRL-21 digital signal processing single microphone (from AliExpress)
- Uses one electret microphone, cleans it up digitally
- Pros:
- Good at cleaning up speech, would take a variety of electret mics
- Cons:
- Really annoying ~80ms processing delay which makes it hard to speak at a normal speed given you can normally hear your own mic in your headphones
- Current draw ~20mA, too high to run off bias supply
?
So I ended up building a variant of the OPA Alice as suggested by Kennjava, but designed to work with the typical 8-20V bias voltage found in aviation radio units:
?
?
?
The mics were knock-off Primo-style unidirection FET-less electrets. One facing the pilot, the other a bit further up the boom arm facing away.??
?
I did originally put a polarity protection diode between the ring and R7, but the extra 0.6V drop was enough to take the OPA1642 down into its noisy territory. Removing this fixed the problem.
?
Certain specs were very hard to dig up:
- Typical General Aviation mic sockets take a PJ-068 5.23mm three pole jack, quite hard to find
- Mic jack pinouts:?
- Tip: "Push to transmit", connect to ground to trigger TX
- Ring: bias voltage + audio input
- Sleeve: connects to radio's ground (radio is then connected to the common grounding point on the aircraft chassis to avoid ground loops)
- Bias voltage in aviation mics: 8-20V
- Max current on bias: 10mA
- Mic input level: 70mV RMS - 1000mV RMS
?
The radio has an input gain which is a bit of a faff to set buried through menus, and as this is a shared aircraft, I didn't want to mess the settings up for everyone else who uses it who tend to use Pooleys headsets. Hence the 10k pot in case I needed to trim my? mic output a bit.
?
I decapitated an old cheap clip-on Maplin pop-filter to borrow its gooseneck and clip, and then put the high impedance part of the circuitry (R3, R4, C1, R1, R2, mics, R6, OPA1642) on a small PCB at the end of the gooseneck, enclosed in aluminium mesh. The mic capsules are mounted to the PCB on strips of foam without blocking the rear holes, with thin enamelled copper wire feeding them to try to mechanically decouple them from the PCB and thus each other. I then ran a shielded 2 core cable up the gooseneck to a second PCB mounted in a small shielded enclosure stuck to the outside of the clip. This second PCB has a finger-friendly rotary knob for the pot. I slid a foam windbreak over the aluminum mesh enclosed PCB to kill plosives.
?
For headphones, I used a pair of Bose QuietComfort SCs. The mic set clips onto the headband just above the left earpiece. I chose the QuietComfort because if the internal battery gets depleted, it just falls back to being a passively noise-cancelled set of headphones.??
?
It works pretty well! Happens to be at the right mic level with the pot mostly towards zero. It is a little tinny, which is to be expected because of how unidirectional capsules are less directional at the bass end, but definitely picks up much less engine noise than the Pooleys headsets. The clip and gooseneck are sturdy and don't flap around at all. There's enough clearance between the headband and my head just above the earphone cup, so that the clip doesn't touch my head at all. The Bose QuietComfort headphones are very comfortable to wear for hours.
My only regret is using a 3.5mm jack socket as the PCB input. This has come out a couple of times, killing the mic, when turning my head to look out the right window if my shoulder has sandwiched the main feed cable between me and the seat.
?
I now have a set of active noise cancelling headphones with microphone for aviation use for <?280, plus the fun of putting it together!
?
|
Hello all.
I found this group while searching for directional microphones. Especially for an headset similar to the one in the above post for the aircraft.
?
On Sat, Dec 21, 2024 at 11:17 PM, <michaeljtbrooks@...> wrote:
It works pretty well! Happens to be at the right mic level with the pot mostly towards zero.
Do you know the mic level you need for the Garmin GNC255? Would be interesting for me to know. Turning your pot mostly towards to zero means, that you are using a low level close to the 70mV you specified?
?
Has anybody ever thought about using an MEMS microphone for this application? Because I think the output impedance might be lower. And there is an MEMS mic called ICS-40800 from TDK with a front and back hole. Regarding the datasheet, it picks up the sound from the front and back side. But mostly ignores the sound from the side (90 degree).
So, I think this is similar to using 2 omni-directional capsules oriented in the opposite? direction. Could be an interesting approach I guess. What do you think?
?
BR
Andreas
|
Le 12/01/2025 à 18:40, aauer1 via groups.io a écrit?: And there is an MEMS mic called ICS-40800 from TDK with a front and back hole. Regarding the datasheet, it picks up the sound from the front and back side. But mostly ignores the sound from the side (90 degree). It's a bi-directional microphone, a.k.a. Figure-8. So, I think this is similar to using 2 omni-directional capsules oriented in the opposite? direction. No, it's not. Two omnidirectional capsules oriented in whatever direction always result in omnidirectionality. What you suggest would be two cardioid mics placed back-to-back and wired out-of-phase. Could be an interesting approach I guess. What do you think? A figure-8 microphone has the same directivity index as a cardioid (4.8dB), so both placed in diffused field receive the same amount of non-desired signal. A hypercardioid mic has a directivity index of 6dB, and rejection is optimal at about 110° on both sides. Sounds from the back are attenuated only by 6dB. It is theoretically the most effective at rejecting ambient noise. However, if the main noise source is at the rear it may be less efficient than a simple cardioid.
|
Le 12/01/2025 à 19:36, Jerry Lee Marcel
via groups.io a écrit?:
EDIT:wo omnidirectional capsules oriented in whatever
direction always result in omnidirectionality.
Two omnidirectional capsules may result in some directivity, but
polarity and level must be adjusted.
What
you suggest would be two cardioid mics placed back-to-back and
wired out-of-phase.
Could be an interesting approach I guess.
What do you think?
A figure-8 microphone has the same directivity index as a cardioid
(4.8dB), so both placed in diffused field receive the same amount
of non-desired signal.
A hypercardioid mic has a directivity index of 6dB, and rejection
is optimal at about 110° on both sides. Sounds from the back are
attenuated only by 6dB.
It is theoretically the most effective at rejecting ambient noise.
However, if the main noise source is at the rear it may be less
efficient than a simple cardioid.
|
And there is an MEMS mic called ICS-40800 from TDK with a front and back hole.?
Regarding this Figure-8 mic, it has an undamped membrane, meaning the frequency response has a +6dB/decade slope. The datasheet is a bit misleading, showing a "Typical Omnidirectional Frequency Response", i.e. with the back port covered. With the back port uncovered, it has this slope, which can be seen in Figure 16 of
?
?
The application note proposes compensating with an electronic filter.
|
Do you know the mic level you need for the Garmin GNC255? Turning your pot mostly towards to zero means, that you are using a low level close to the 70mV you specified?
?
Sorry, I meant I turned the pot mostly towards zero resistance.
I've not measured the output amplitude properly during typical use with the foam on, but I'd shot for about 300mV peak-to-peak at zero pot resistance on the oscilloscope to close proximity normal speech when it was in pieces on my bench. The Garmin GNC255 requires 70mV - 1000mV depending on what its mic input gain has been set to.
?
There's plenty of scope to increase the amplitude on my mic setup. The second op-amp stage is just a buffer so could easily be given gain via a feedback resistor.
For reference, the Bose A20 mic outputs 600mV at 114dB SPL.
?
?
|
Timothy Aguana
