Discussion:
MEMS SNR Specifications
(too old to reply)
Ralph Jones
2018-08-17 21:55:36 UTC
Permalink
Raw Message
There’s been some interesting discussion here about Micro-ElectroMechanical Systems (MEMS) microphones for ambisonic mic arrays. These devices seem to offer some compelling qualities (particularly small size and low cost) but their signal-to-noise spec is 65 dBA. Some folks posting here have seemed to suggest that this level of noise might possibly be acceptable.

IMHO, based on decades of experience, a microphone with a signal-to-noise ratio of 65 dBA is useless for professional recording. That's in the SNR range typical of consumer cassette tape machines or analog AM radio.

For comparison, consider that professional large-diaphragm condenser mics achieve 120-130 dB SNR. The Sennheiser Ambeo has an SNR of about 110 dB. Portable digital audio recorders and popular audio interfaces make about 100 dB, and Red Book CDs 98 dB. Even vinyl records are about 6 dB quieter than MEMS mic elements.

Put a MEMS mic at the input of a digital recorder and you’re wasting 35 dB of dynamic range (not to mention the case of using several of them in an ambisonic array). That’s huge. A symphony orchestra playing at mezzoforte or louder may mask the noise if the mic is reasonably close to the stage, but forget trying to record softer passages. Recordists trying to capture natural ambiences will be sorely disappointed; a lot of what they’re trying to record will simply disappear into the noise floor. You might get away with using it for non-critical functions like background crowd noise for telecasts of sporting events, but that’s about it.

MEMS mics appear to me to have been developed for applications that require very small size, physical ruggedness, reasonable frequency response and low cost, but can tolerate a high noise floor. There are lots of such use cases, but professional audio is not one of them. If we wish to advance the art of ambisonic recording and reproduction, we cannot compromise basic performance specifications for the sake of a trendy idea.

Ralph Jones
Paul Hodges
2018-08-17 22:56:21 UTC
Permalink
Raw Message
--On 17 August 2018 14:55 -0700 Ralph Jones <***@comcast.net>
wrote:

> Some folks posting here have seemed to suggest that this level of
> noise might possibly be acceptable.

Well, firstly we don't know the actual specification of the devices
used by Zylia. And secondly, using an array of nineteen to generate an
output gives the possibility of significant improvement, because the
sound source signals are correlated and the noise is uncorrelated.

How this holds up in practice at higher orders and higher frequencies I
will attempt to judge when I get my hands on the ZM-1 rather than just
predicting failure in advance (which is not consistent with the reviews
I've seen heard and read).

Paul

--
Paul Hodges
Jack Reynolds
2018-08-17 23:37:22 UTC
Permalink
Raw Message
Are you sure the Ambeo has 110dB SNR?

Sent from my iPhone

> On 17 Aug 2018, at 23:56, Paul Hodges <pwh-***@cassland.org> wrote:
>
> --On 17 August 2018 14:55 -0700 Ralph Jones <***@comcast.net>
> wrote:
>
>> Some folks posting here have seemed to suggest that this level of
>> noise might possibly be acceptable.
>
> Well, firstly we don't know the actual specification of the devices
> used by Zylia. And secondly, using an array of nineteen to generate an
> output gives the possibility of significant improvement, because the
> sound source signals are correlated and the noise is uncorrelated.
>
> How this holds up in practice at higher orders and higher frequencies I
> will attempt to judge when I get my hands on the ZM-1 rather than just
> predicting failure in advance (which is not consistent with the reviews
> I've seen heard and read).
>
> Paul
>
> --
> Paul Hodges
>
> _______________________________________________
> Sursound mailing list
> ***@music.vt.edu
> https://mail.music.vt.edu/mailman/listinfo/sursound - unsubscribe here, edit account or options, view archives and so on.
Bo-Erik Sandholm
2018-08-18 07:41:13 UTC
Permalink
Raw Message
According to the document linked to below that relates self noise values to
real world applications 110 SNR cannot be related to the commonly used
reference sound level.
110 dBA SNR would be 16 dB below absolute quiet.

If the value 70dBA that I found for the infineon dual membrane MEMS mic is
related to 1 Pascal, then it's self noise is around 24 dB which is not
strictly studio quality.
But not really horrible.
If it is related to max 10% distortion which is at 135 dBA thats not a
realistic comparison value as the result is a self noise of 65 dBA.

That would be a noise source not a microphone :-) !

So a bit of apples and oranges comparison is going on 😎


http://www.neumann.com/homestudio/en/what-is-self-noise-or-equivalent-noise-level

SIGNAL-TO-NOISE RATIO
Another way to document the noise performance is to specify the
signal-to-noise ratio. But relative to what signal? The reference sound
pressure level for noise measurements is 94 dB (which equals a sound
pressure of 1 pascal). So you can simply calculate:

Signal-to-noise (db-A) = 94 dB – self-noise (dB-A)



The actual signal-to-noise ratio in use, of course, depends on the sound
pressure level of your sound source.


Bo-Erik

On Sat, 18 Aug 2018 01:37 Jack Reynolds, <***@gmail.com> wrote:

> Are you sure the Ambeo has 110dB SNR?
>
> Sent from my iPhone
>
> > On 17 Aug 2018, at 23:56, Paul Hodges <pwh-***@cassland.org> wrote:
> >
> > --On 17 August 2018 14:55 -0700 Ralph Jones <***@comcast.net>
> > wrote:
> >
> >> Some folks posting here have seemed to suggest that this level of
> >> noise might possibly be acceptable.
> >
> > Well, firstly we don't know the actual specification of the devices
> > used by Zylia. And secondly, using an array of nineteen to generate an
> > output gives the possibility of significant improvement, because the
> > sound source signals are correlated and the noise is uncorrelated.
> >
> > How this holds up in practice at higher orders and higher frequencies I
> > will attempt to judge when I get my hands on the ZM-1 rather than just
> > predicting failure in advance (which is not consistent with the reviews
> > I've seen heard and read).
> >
> > Paul
> >
> > --
> > Paul Hodges
> >
> > _______________________________________________
> > Sursound mailing list
> > ***@music.vt.edu
> > https://mail.music.vt.edu/mailman/listinfo/sursound - unsubscribe here,
> edit account or options, view archives and so on.
> _______________________________________________
> Sursound mailing list
> ***@music.vt.edu
> https://mail.music.vt.edu/mailman/listinfo/sursound - unsubscribe here,
> edit account or options, view archives and so on.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.music.vt.edu/mailman/private/sursound/attachments/20180818/40e5ff31/attachment.html>
Chris Woolf
2018-08-18 08:56:57 UTC
Permalink
Raw Message
I think there is indeed some confusion in this discussion between the
signal-to-noise ratio of these mics, and dynamic range.

The first is conventionally related to 1Pa/94dB SPL, and one then needs
to add in a Max SPL figure to get the dynamic range.

We need both bits of information to understand the practicality of any mic.

A noise floor of 24dBA (related to 1Pa) is about par for a small
personal electret mic. A dynamic range of >115dB is what one would wish
for in decent professional mics - that would be a noise floor of 15dBA
and a max SPL of >130dB (with a distortion figure of 3 or 5%).

Chris Woolf (ex editor of Microphone Data)


On 18/08/2018 08:41, Bo-Erik Sandholm wrote:
> According to the document linked to below that relates self noise values to
> real world applications 110 SNR cannot be related to the commonly used
> reference sound level.
> 110 dBA SNR would be 16 dB below absolute quiet.
>
> If the value 70dBA that I found for the infineon dual membrane MEMS mic is
> related to 1 Pascal, then it's self noise is around 24 dB which is not
> strictly studio quality.
> But not really horrible.
> If it is related to max 10% distortion which is at 135 dBA thats not a
> realistic comparison value as the result is a self noise of 65 dBA.
>
> That would be a noise source not a microphone :-) !
>
> So a bit of apples and oranges comparison is going on 😎
>
>
> http://www.neumann.com/homestudio/en/what-is-self-noise-or-equivalent-noise-level
>
> SIGNAL-TO-NOISE RATIO
> Another way to document the noise performance is to specify the
> signal-to-noise ratio. But relative to what signal? The reference sound
> pressure level for noise measurements is 94 dB (which equals a sound
> pressure of 1 pascal). So you can simply calculate:
>
> Signal-to-noise (db-A) = 94 dB – self-noise (dB-A)
>
>
>
> The actual signal-to-noise ratio in use, of course, depends on the sound
> pressure level of your sound source.
>
>
> Bo-Erik
>
> On Sat, 18 Aug 2018 01:37 Jack Reynolds, <***@gmail.com> wrote:
>
>> Are you sure the Ambeo has 110dB SNR?
>>
>> Sent from my iPhone
>>
>>> On 17 Aug 2018, at 23:56, Paul Hodges <pwh-***@cassland.org> wrote:
>>>
>>> --On 17 August 2018 14:55 -0700 Ralph Jones <***@comcast.net>
>>> wrote:
>>>
>>>> Some folks posting here have seemed to suggest that this level of
>>>> noise might possibly be acceptable.
>>> Well, firstly we don't know the actual specification of the devices
>>> used by Zylia. And secondly, using an array of nineteen to generate an
>>> output gives the possibility of significant improvement, because the
>>> sound source signals are correlated and the noise is uncorrelated.
>>>
>>> How this holds up in practice at higher orders and higher frequencies I
>>> will attempt to judge when I get my hands on the ZM-1 rather than just
>>> predicting failure in advance (which is not consistent with the reviews
>>> I've seen heard and read).
>>>
>>> Paul
>>>
>>> --
>>> Paul Hodges
>>>
>>> _______________________________________________
>>> Sursound mailing list
>>> ***@music.vt.edu
>>> https://mail.music.vt.edu/mailman/listinfo/sursound - unsubscribe here,
>> edit account or options, view archives and so on.
>> _______________________________________________
>> Sursound mailing list
>> ***@music.vt.edu
>> https://mail.music.vt.edu/mailman/listinfo/sursound - unsubscribe here,
>> edit account or options, view archives and so on.
>>
> -------------- next part --------------
> An HTML attachment was scrubbed...
> URL: <https://mail.music.vt.edu/mailman/private/sursound/attachments/20180818/40e5ff31/attachment.html>
> _______________________________________________
> Sursound mailing list
> ***@music.vt.edu
> https://mail.music.vt.edu/mailman/listinfo/sursound - unsubscribe here, edit account or options, view archives and so on.


---
This email has been checked for viruses by Avast antivirus software.
https://www.avast.com/antivirus
Fons Adriaensen
2018-08-18 09:52:59 UTC
Permalink
Raw Message
On Sat, Aug 18, 2018 at 09:41:13AM +0200, Bo-Erik Sandholm wrote:

> According to the document linked to below that relates self noise values to
> real world applications 110 SNR cannot be related to the commonly used
> reference sound level.
> 110 dBA SNR would be 16 dB below absolute quiet.

Maybe possible with an unexistium membrane and left twisting electrons ... :-)

> If the value 70dBA that I found for the infineon dual membrane MEMS mic is
> related to 1 Pascal, then it's self noise is around 24 dB which is not
> strictly studio quality. But not really horrible.

It's the self noise figure which matters in the end. It simply means
that if you record a sound with an SPL of say 60 dB (at the mic), the
S/N ratio would be 60 - 24 = 36 dB. Using a 'pro' mic with a self
noise of 10 dB the S/N ratio would be 50 dB. In both cases assuming
the preamp doesn't add a significant amount of noise.

Noise of 'virtual mics' (or ambisonice components) synthesised from
multiple capsules is not so simple.

If S_i is the signal from the i-th capsule then the beam signal is
of the form

S_beam = sum (a_i * S_i) i = 1..B

where the a_i are complex gains depending of frequency. Now if the
a_i are scaled such that the on-axis sensitivity of the beam is the
same as that for a single capsule, then the noise power in the beam
signal is the sum of the squares of the a_i:

N_beam = sum (a_i ^ 2)

For an 'omni' pattern this typically results in a significant
improvement, up to 3dB * log2 (N) in the best case (ommi
capsules at low and medium frequencies).

For directive patterns, almost anything can happen depending
on if the capsules are omni or directional, if they are in
free space or on a solid body, and the order of the pattern.
The actual noise will not be white which means that traditional
noise figures (e.g. A-weighted) become more or less useless.

Typically you may see some improvement over the single capsule
noise level in some small frequency region, and more noise
in other frequency bands.

For some examples, get hold of the original Eigenmic manual
which shows noise performance as a function of frequency
for each of the available patterns. That's the only example
I know of a manufacturer documenting this, most (including
Zylia) will hide the truth in lots of blabla.

Ciao,

--
FA
Stefan Schreiber
2018-08-18 17:08:00 UTC
Permalink
Raw Message
Many thanks, the information below was very helpful. (At least
speaking for myself.)

So I suspect that there are some performance issues involved.
(However, the ZM-1 prototype specs  are certainly not that
“amateurish” as some have suggested...)

In the end the ZM-1 is not really on the market yet. All we have is
some “anecdotical evidence” about the microphone’s prototype
performance.

Current product state:
http://www.zylia.co/purchase.html

(“ZYLIA is available in a pre-order”)

Otherwise this microphone and its sw has been reviewed in many places.

Just one link:

https://www.soundonsound.com/reviews/zylia-zm-1

A fair review (as cited) would take note of the microphone’s many uses
and high flexibility.

Best,

Stefan Schreiber

- - -

Citando Chris Woolf <***@chriswoolf.co.uk>:

> I think there is indeed some confusion in this discussion between
> the signal-to-noise ratio of these mics, and dynamic range.
>
> The first is conventionally related to 1Pa/94dB SPL, and one then
> needs to add in a Max SPL figure to get the dynamic range.
>
> We need both bits of information to understand the practicality of any mic.
>
> A noise floor of 24dBA (related to 1Pa) is about par for a small
> personal electret mic. A dynamic range of >115dB is what one would
> wish for in decent professional mics - that would be a noise floor
> of 15dBA and a max SPL of >130dB (with a distortion figure of 3 or
> 5%).
>
> Chris Woolf (ex editor of Microphone Data)
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.music.vt.edu/mailman/private/sursound/attachments/20180818/f02eaeef/attachment.html>
Bo-Erik Sandholm
2018-08-18 20:33:15 UTC
Permalink
Raw Message
My last comment on MEMS performance.

One piece price from Mouser 1.80 Euro.

Complete data sheet here with graphs...

https://www.mouser.se/ProductDetail/Infineon/IM69D130V01XTSA1/?qs=%2fha2pyFaduj4xwaVFOJc2igDZAeL7qWZE4cyVm2vU0DkFu5Rl9gaWnMlZsUg39Pw&utm_source=octopart&utm_medium=aggregator&utm_campaign=726-IM69D130V01XTSA1&utm_content=Infineon

On Sat, 18 Aug 2018 19:08 Stefan Schreiber, <***@mail.telepac.pt> wrote:

> Many thanks, the information below was very helpful. (At least
> speaking for myself.)
>
> So I suspect that there are some performance issues involved.
> (However, the ZM-1 prototype specs are certainly not that
> “amateurish” as some have suggested...)
>
> In the end the ZM-1 is not really on the market yet. All we have is
> some “anecdotical evidence” about the microphone’s prototype
> performance.
>
> Current product state:
> http://www.zylia.co/purchase.html
>
> (“ZYLIA is available in a pre-order”)
>
> Otherwise this microphone and its sw has been reviewed in many places.
>
> Just one link:
>
> https://www.soundonsound.com/reviews/zylia-zm-1
>
> A fair review (as cited) would take note of the microphone’s many uses
> and high flexibility.
>
> Best,
>
> Stefan Schreiber
>
> - - -
>
> Citando Chris Woolf <***@chriswoolf.co.uk>:
>
> > I think there is indeed some confusion in this discussion between
> > the signal-to-noise ratio of these mics, and dynamic range.
> >
> > The first is conventionally related to 1Pa/94dB SPL, and one then
> > needs to add in a Max SPL figure to get the dynamic range.
> >
> > We need both bits of information to understand the practicality of any
> mic.
> >
> > A noise floor of 24dBA (related to 1Pa) is about par for a small
> > personal electret mic. A dynamic range of >115dB is what one would
> > wish for in decent professional mics - that would be a noise floor
> > of 15dBA and a max SPL of >130dB (with a distortion figure of 3 or
> > 5%).
> >
> > Chris Woolf (ex editor of Microphone Data)
> -------------- next part --------------
> An HTML attachment was scrubbed...
> URL: <
> https://mail.music.vt.edu/mailman/private/sursound/attachments/20180818/f02eaeef/attachment.html
> >
> _______________________________________________
> Sursound mailing list
> ***@music.vt.edu
> https://mail.music.vt.edu/mailman/listinfo/sursound - unsubscribe here,
> edit account or options, view archives and so on.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.music.vt.edu/mailman/private/sursound/attachments/20180818/fc7e9c15/attachment.html>
Steven Backer
2018-08-18 23:20:03 UTC
Permalink
Raw Message
Following up on the insightful comments from Fons et al. -

The increase or decrease in noise relative to the self-noise of a single mic is often referred to in the literature as White Noise Gain (WNG). This can be calculated or measured for linear filtering operations such as Ambisonics B-format encoding or virtual microphone beamforming. Lower numbers are worse; negative values indicate an increase in noise level relative to a single omni mic, and positive values indicate a reduction in relative noise level.

For the Eigenmike, we take this into account as a design constraint in our software, and even give the user a few options to trade off this constraint against the operating regions of the higher-order signals. We did post some documents specifying the WNG (as a function of frequency) on our website not too long ago, both for the B-format encoding (“Eigenbeams”) as well as the virtual microphones (“Modal Beamformers”). This is specific to our hardware and software and may differ in others’ implementations.

Eigenbeams:
https://mhacoustics.com/sites/default/files/Eigenbeam%20Datasheet_R01A.pdf <https://mhacoustics.com/sites/default/files/Eigenbeam%20Datasheet_R01A.pdf>
Modal Beamformer:
https://mhacoustics.com/sites/default/files/Beamformer%20Datasheet_R02A.pdf <https://mhacoustics.com/sites/default/files/Beamformer%20Datasheet_R02A.pdf>

-Steven

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.music.vt.edu/mailman/private/sursound/attachments/20180818/cd3009ac/attachment.html>
Marc Lavallée
2018-08-19 00:02:37 UTC
Permalink
Raw Message
Interesting graphs; eigenbeams with higher orders have more low
frequency noise, without much impact on beam forming.

Marc

Le 18/08/2018 à 19:20, Steven Backer a écrit :
> Following up on the insightful comments from Fons et al. -
>
> The increase or decrease in noise relative to the self-noise of a single mic is often referred to in the literature as White Noise Gain (WNG). This can be calculated or measured for linear filtering operations such as Ambisonics B-format encoding or virtual microphone beamforming. Lower numbers are worse; negative values indicate an increase in noise level relative to a single omni mic, and positive values indicate a reduction in relative noise level.
>
> For the Eigenmike, we take this into account as a design constraint in our software, and even give the user a few options to trade off this constraint against the operating regions of the higher-order signals. We did post some documents specifying the WNG (as a function of frequency) on our website not too long ago, both for the B-format encoding (“Eigenbeams”) as well as the virtual microphones (“Modal Beamformers”). This is specific to our hardware and software and may differ in others’ implementations.
>
> Eigenbeams:
> https://mhacoustics.com/sites/default/files/Eigenbeam%20Datasheet_R01A.pdf <https://mhacoustics.com/sites/default/files/Eigenbeam%20Datasheet_R01A.pdf>
> Modal Beamformer:
> https://mhacoustics.com/sites/default/files/Beamformer%20Datasheet_R02A.pdf <https://mhacoustics.com/sites/default/files/Beamformer%20Datasheet_R02A.pdf>
>
> -Steven
>
> -------------- next part --------------
> An HTML attachment was scrubbed...
> URL: <https://mail.music.vt.edu/mailman/private/sursound/attachments/20180818/cd3009ac/attachment.html>
> _______________________________________________
> Sursound mailing list
> ***@music.vt.edu
> https://mail.music.vt.edu/mailman/listinfo/sursound - unsubscribe here, edit account or options, view archives and so on.
Stefan Schreiber
2018-08-19 03:01:27 UTC
Permalink
Raw Message
“My last comment on MEMS performance.

One piece price from Mouser 1.80 Euro.

Complete data sheet here with graphs...

https://www.mouser.se/ProductDetail/Infineon/IM69D130V01XTSA1/?qs=%2fha2pyFaduj4xwaVFOJc2igDZAeL7qWZE4cyVm2vU0DkFu5Rl9gaWnMlZsUg39Pw&utm_source=octopart&utm_medium=aggregator&utm_campaign=726-IM69D130V01XTSA1&utm_content=Infineon



(Bo-Erik)
- - - -

Citando Bo-Erik Sandholm <***@gmail.com>:My last comment on MEMS
performance.One piece price from Mouser 1.80 Euro.

(Ammendment) Manufacturer’s data sheet:

https://www.infineon.com/dgdl/Infineon-IM69D130-DS-v01_00-EN.pdf?fileId=5546d462602a9dc801607a0e46511a2e

(If Zylia even uses this relatively new MEMS microphone, by the way...)

< Dynamic range of 105dB >

Best regards

Stefan
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.music.vt.edu/mailman/private/sursound/attachments/20180819/fde543d1/attachment.html>
Loading...