Compression lesson

[Mary Beth]

Okay gang,

I need to learn more about compression options in the Sonnet processors. Now just by being a CI, the processor has to compress a wide dynamic range into a small electrical dynamic range. I've read about that. What I would like to learn about are the setting options our audiologists have that relate to compression. I know they can change the compression ratio (how much change in input results in 1 unit of change in output). So I believe I am presently at 3:1. If I go lower than 3, there would be less compression so would that mean that the CI would not dampen steady noise sounds as much? Can the audiologists change a threshold kneepoint in some way? Can they change at what loudness, this compression triggers? Can the audiologists change the attack time and release time for compression?

I'd appreciate reading anything you have learned along the way about compression and our processors.

Thanks for your help,

Mary Beth

[Ivana Marinac]

Very Interesting question Mary Beth.

Unfortunately, I can't help you.

[Adam]

Marybeth

I have done a great of reading on this, early on in my CI journey. I will have to revisit it as I do not want to give wrong information. Will touch base soon

[Adam]

Marybeth

You might already know this but this is some info dealing with the automatic gain control and compression

Automatic Gain Control Provides

a Carefree Listening Experience

The Challenge

One of the biggest challenges facing any cochlear implant system is that of compressing a widely varying loudness spectrum into a very small electrical dynamic range. The ear with normal hearing can generally process loudness inputs from 0 to 120 dB HL with little difficulty. Of course, as we approach our loudness "limit," we begin to experience negative effects, like ear discomfort or even pain. These cues tell us that the sound we are hearing is loud enough to cause permanent damage to our hearing, and they prompt us to somehow change the environment to protect our ears.

With electrical hearing, the picture is very different. Here, the ear with an implant generally has a dynamic range of 10-20 dB, and sometimes it is even as low as just a few decibels. This is a result of the processes that created the sensorineural hearing loss in the first place, and also of the way in which nerve fibres respond to electrical stimulation. All of this means that the cochlear implant system needs to be able to accurately represent subtle loudness cues, as rapidly as they change in the environment, across the wide range of normal hearing by compressing those cues into the very narrow dynamic range of electrical hearing. It is the subtle differences between energy peaks in a speech signal that let us decode that signal and give it meaning, so we need to represent those cues accurately, but also as rapidly as the subtle changes in running speech, or in a musical piece.

Note the many rapid loudness variations as the word “cochlea” is spoken.

Possible solutions

A system could, of course, use only a single-stage compression scheme – but the hearing aid literature tells us that speech understanding with this type of system is rather poor. It’s poor because the system is trying to do two things in one step: while it is trying to compensate for the overall volume of the input signal (which has to happen relatively slowly), it can’t react quickly enough to sudden changes in the sound level. The result is that the user can perceive “dead points” or moments when sound cuts out completely. Therefore, a single-stage compression scheme will not be sufficiently able to manage overall sound level while still accurately representing subtle loudness cue changes that happen very rapidly.

The best solution is a two-stage compression design. The first stage, the dynamic step, is the one we will focus on here. This first stage is the Dual-Loop Automatic Gain Control (AGC). The second stage, which occurs later in the sound coding process, is a static step called maplaw. The maplaw is a compression parameter that is applied equally to all channels and determines where “soft,” “medium,” and “loud” sounds are mapped into the user's individually-measured dynamic range. The maplaw component of the coding strategy is customisable by the audiologist during the fitting session.

The dual-loop AGC – the first, dynamic, stage of the process – is a clever approach to accurately representing loudness variations and details in constantly-changing environments. Although the audiologist will often test a patient’s ability to hear in background noise, this test environment, in a sound-treated booth with a fixed presentation level for speech and a fixed level for noise, is not very representative of the real world. MED-EL’s dual-loop AGC, however, uses two monitoring systems to track and manage incoming sound based on the overall sound level – making the MED-EL system automatically adaptive to the environment.

How does dual-loop AGC work?

AGC systems all have one shared problem. They need to manage the overall sound level, but they also need to manage unexpected intense transient sounds in order to remain comfortable to the user. Those two things require completely different approaches. Overall sound level needs to be monitored slowly, but transient sounds come and go so quickly that a system with just one loop would miss them completely, resulting in uncomfortably loud stimulation for the user every time an unexpectedly loud sound occurs in the environment. If the AGC is built to react quickly, then the user perceives the sound “pumping in and out.” There is a fine balance between what type of attack time is needed vs. what can be perceived, and what type of release time is needed vs. what can be perceived, and the two can’t easily coexist peacefully in one circuit.

MED-EL resolves this paradox in the following way. All incoming sound passes through two peak detectors that co-exist side-by-side. One detector analyses rather slowly (on the order of several hundred milliseconds). The other analyses quickly. You can almost imagine the slow detector analysing at a relatively slow “rate,” checking in now and then to see what the overall level is. The fast detector is analysing very quickly – it is looking for loud, transient sounds.

Both of these detectors have the ability to change the gain of the microphone. Usually the slow detector is in control (even though both are always working). So if the overall sound level of the environment changes (as when walking from a quiet room to a noisy room), the slow detector adjusts the microphone gain to be sure the input is within the sound processing window of the implant system. But if the fast detector notices a sound that exceeds the slow detector by more than 6 dB in a very short time period, it quickly takes control, reduces the gain of the microphone (if necessary), and then gives control back to the slow detector. The slow detector then just picks up right where it left off - because the overall sound level didn’t really change. This allows the management of the transient sound to be handled without the user noticing any negative effects.

Taking a closer look

First, let’s look at the slow detector. The slow detector operates essentially as an automatic volume control. MED-EL implant users have the ability to control volume with an adjustment button on the processor or with the Fine Tuner (depending on the processor style). However, nearly all patients report that they do not feel the need to adjust volume as they go about their day, regardless of the changing environments that they experience. That’s because the slow detector is constantly focusing on the sound level in the present moment. There is no perception of anything adjusting in the background – just effortless movement between one environment and another without the patient being required to make adjustments manually to support the processor.

The slow detector is in control the majority of the time, until a loud transient sound is detected by the system. At this point, control shifts to the fast detector. The fast detector’s job is to identify the transient, compress it very rapidly so that the patient doesn’t experience discomfort, and then hand control immediately back to the slow detector.

The end result of this process is shown here (above). On the first line, you see the original sound signal as it changes over time. It is a speech signal, and you’ll notice a loud sound obscuring the speech signal in the first moments of the tracing. The middle line shows the result of a standard single-loop compression system. The transient sound is identified and rapidly compressed, but the system must take a relatively long time to release the compression to avoid having the patient hear a “motor-boating” or “pumping” sound. In fact, it appears that the speech signal doesn’t return to its normal loudness level until more than a full second has passed.

The third line, however, shows the result of a dual-loop AGC. The transient sound is managed nearly instantaneously, and the management of the overall sound level is returned back to the slow detector within just a fraction of a second.

The result

The end result for the patient is unique – the patient just puts on the system and goes about their day. The dual-loop AGC manages the varying environments that the patient encounters, without manual input, over a 75 dB input range (accurately processing inputs from 25 – 100 dB SPL). In the patient’s sound experience, soft sounds are appropriately soft and loud sounds are proportionally loud, but specific program or setting changes to accommodate the differences between noisy environments, music, quiet environments or speech as the primary signal are simply not necessary.

The dual-loop AGC design increases the robustness of the signal when high level transient sounds are present, which means a better chance at better speech understanding for the user, even in the most adverse listening conditions. To view a study that attempted to recreate the varying listening conditions found in the real world and assess the performance of the front-end preprocessing of all three cochlear implant systems in this setting, click here.

Adam

[hadron]

Mary Beth,

Good question. Sorry for late reply as I have been in Geneva at CERN.

The compression/kneepoint settings are part of the Maestro system software used to map your processor. It is used for all MED-EL processors which can be programmed by Maestro.

I have not seen the user interface screens for the Maestro 6.0 mapping software but I am familiar with the Maestro 4.0 screens. The ASM screen allows you to set the AGC compression ratio to one of four values (2.0:1, 2.5;1, 3.0:1, 3.5:1). The default is 3.0:1. The compression ratio is usually lowered if loud sounds are too soft (e.g. music) Which increases the difference between loud and soft sounds. If loud sounds are uncomfortable, the compression ratio can be increased to 3.5:1 which decreases the difference between soft and loud sounds.

For kneepoints I believe it you look at the maplaw screen and click on the frequency band tab you will be able to set your values.

When you next visit your audiologist ask her if she could show you the Maestro 6.0 mapping software screens she used to map your SONNET.

What I discussed is presented in more detail in the book "Programming Cochlear Implants" second edition 2014 by Wolfe and Schaefer. I also bought the first edition in 2010 and highly recommend the book. It is used as a textbook in the audiology graduate program for many universities.

Wolfe and Schaefer have done a lot of research concerning hearing in noise with cochlear implants. They have analyzed almost every ALD which interfaces with a cochlear implant. Their latest research discusses the use of the Roger Pen with cochlear implants. Their analysis shows that CI users have the potential to hear significantly better in noise if they use ALDs.

[Mary Beth]

Thanks Hadron! I will study up on this a bit. I know I am at the default 3.0:1 compression ratio now. I do not know what the kneepoint is set at so I will ask next month. I am not having any compression issues with the Opus2 and Rondo. However the Sonnet is a bit odd when presented with low frequency steady noise. I think the attack time is too fast and the release time is too slow. However, I'm in the US and so I only have one mic active right now and wonder if this will fix itself when the dual mics are turned on.

Thanks again for your help

[Kara of Canada]

Yes I just had my mapping. She adjusted the compression ( nit sure to what degree) and there is a setting for the adaptive mode where it can cap the signal input and to what degree. I had that adjusted to make it more agressive in the loud settings so it will make sound more comfortable sooner. It was at 70% but she bumped it to 80% which it caps more agressively. Hope this helps others are more technical!!!

[hadron]

Mary Beth,

Now that you mention it I did notice that when I am wearing the Sonnet and start the washing machine, which gives off this low steady state frequency noise, the processor starts acting strangely and when I try to talk I can barely hear myself. Once I step away from the washing machine things return to normal.

No such problem with the Opus 2.

[hadron]

Kara,

Sounds like your audiologist took good care of you. Glad to hear.

If you haven't consider sending a thank you email to your audiologist. I believe it helps grow your relationship. I always follow up an appointment with my surgeon or audiologist with a quick thank you email. I now email my surgeon whenever I have a problem and he promptly replies.

[Mary Beth]

Hadron,

Another long time Opus2 user who switched to the Sonnet is having exactly the same problem as us. He says it makes him think his batteries are dying.

Since all three of us are in the US and awaiting the dual mic approval, do you think it will fix itself when both mics are active? It does this when I am in the car with the heater on high and I can hardly hear my own voice.

I want to figure this out because in every other way, the Sonnet is great. However this is a problem.

If by chance, you find a solution please let me know.

Thanks,

Mary Beth

[Kara of Canada]

Thanks Hadron. That's a good idea.

[Ivana Marinac]

Mary Beth, why wouldn't you put this question directly?

[Mary Beth]

Hi Ivana,

I did talk with my audiologist about the problem last week at my mapping. She contacted her clinical account manager who first told her no one reported the same problem but then came back and said that there are ways to address it however that part of the software has not been unlocked in the US yet. She suggested we either address compression or sensitivity in the meantime. After discussing it with my audiologist, we decided to wait one more month until my next mapping to see what happened in the meantime. We froze my lowest electrodes though as 250 Hz was a bit high and this compression issue is triggered by a low frequency static sound.

In the meantime I have been learning about CIs and compression and getting more comfortable with the terms involved like compression ratio, kneepoint, attack time and recovery time etc.

One gentleman who is experiencing the same difficulty, was following up with the in-house audiologists in North Carolina. I am waiting to find out what he learns. He knows those audiologists.

Hadron is the third person who has experienced the same thing that I know of.

I have also asked Erik for help with this from the clinical managers at MedEl's center.

Since I am not aware of anyone reporting this issue who has the dual mics activated, I am wondering if this indeed is a temporary problem that will fix itself once the US is allowed to have the dual mics turned on.

So I guess you can say that I am in the information gathering stage. Smile.

It's like a puzzle and right now we seem to be missing a piece.

Mary Beth

[Ivana Marinac]

Pardon me Mary Beth,

I was not aware that you have contacted Med-El, that is why I have asked the aforementioned question. But my intention was directed toward getting possible opinions from the other markets: the European more specifically. 

But now I see that you are in safe hands...

[Adam]

Sounds like you are headed down the right path.

[Adam]

I came very close to buying that book after reading some of it. I believe there is another out there but the name escapes me at the moment

[Mary Beth]

Ivana,

No worries. I did not take any offense by your post. I am in good hands. Erik is on top of it. I will let you know what I find out. I always appreciate your comments.

Mary Beth

[Adam]

I posed this question to a freind who works for Medel and has her doctorate. Hadron explained quite well but thought I would post the reply. When she replied, she didn't know you had already seen your Audi.

First and foremost the Sonnet does NOT work like a hearing aid. All of her questions come from knowing something about how hearing aids work. It happens a lot. The attack and release times cannot be adjusted (and you wouldn't want to). There is an interaction of the knee point on M levels based on changes to sensitivity. The knee point is not a feature that can be changed by thresholds. Lowering the compression ratio to 2:1 will work as she asks - it will not dampen noise as much. There is another form of compression in our processor called maplaw. The higher the number the more that soft sounds are pushed up into the input dynamic range.

Honestly, she should be having this conversation with her audiologist.

[Ivana Marinac]

Ivana,

No worries. I did not take any offense by your post. I am in good hands. Erik is on top of it. I will let you know what I find out. I always appreciate your comments.

Mary Beth

 

Tnx Mary Beth,

 

I am quite interesting for each one's observation - I will definitely buy this book in order to understand better basic theory behind a real product. 

Keep tracking your work - it's quite impressed way you are dealing with observations.

[Mary Beth]

Adam,

Thanks for reaching out and asking my questions to your friend at MedEl. This is definitely an area that I need to learn more about. It is helpful to have learned these things from your post:

-attack and release times can not be set

-kneepoint changes with sensitivity changes (not threshold changes)

-lowering compression ratio will not dampen noises as much

-MAPLAW is another form of compression (the higher the number

the more that soft sounds are pushed up into the input dynamic

range)

I appreciate your help.

[Mary Beth]

Hi Hadron,

The other person went to his audiologist about the same issue with the Sonnet. His audiologist made these adjustments:

Changed compression ratio from default of 3.0:1 to 2.5:1 and then bumped up electrode 1 a click or two to compensate for the loss of low quiet sounds.

Just wanted to share what is being tried.

I will touch base with him in a few weeks to see if it helped. I return for another mapping on February 8th so I may make changes at that time. However, my audiologist and I took a different route about this issue at my last mapping a week ago. We slightly lowered the first two electrodes and left the compression ratio at 3.0:1. There is a noticeable improvement.

I have located the textbook (borrowing it from an audiologist friend) and I will study up!

Mary Beth

[hadron]

Mary Beth,

Appreciate the update.

I like the approach you used.

Appreciate both of you trying to find a fix.

You will enjoy the book. The 2014 edition is 3 times bigger than the 2010 first edition. I have exchanged emails with Schaefer and she is highly motivated. Big on advancing CI aural rehabitation in adults and improving hearing ability in noise.

[Mary Beth]

Hadron,

Which state do you live in? I am in New York.

Mary Beth

[hadron]

Hi Mary Beth,

 

I split my time between Florida and Massachusetts. Since it is winter I am currently in Florida. I just got back from a few weeks in Geneva, Switzerland so it is nice to be in Florida. However it has been a cold month lately in Florida due to El Nino.

[Mary Beth]

That sounds lovely! Is your audiologist in Florida or Mass or do you work with one in each location?