LFE in a stereo downmix

[KMO]

Both Dolby and ITU-R BS.775-3 specify -3dB for the center and side channels when downmixing.

So Dolby does. I'd never noticed that about the "Lo/Ro" downmix. I've only ever paid attention to the "Lt/Rt" Pro Logic-encoding downmix, which doesn't lower the surrounds like that. Rather it's -1.2dB to the same side and -6.2dB to the other side, plus phase shenanigans. I guess you want to do either phases shenanigans or level lowering to de-emphasise the stuff that's supposed to be behind you.

 

[dasdoing]

That post you're quoting is nonsense, at least with respect to the specific LFE bit you're relying on.

Home systems always have +10dB for the LFE out of the box, and should do for standard Dolby/DTS mixes. The reference 0dBFS for the LFE is 115dB(C), compared to 105dB(C) for the mains, same as the cinema. (You don't usually play the entire system at reference in a small non-cinema like room, but that doesn't affect the point about the 10dB LFE<>main offset).

However, he is right about the surround offset difference - home systems are calibrated for equal-level mains, unlike the cinema, and that should be allowed for when preparing the consumer mix.

digitalfrost's mix calculation looks reasonable to me. Including the detail of assuming 3dB non-coherent summing for the full-bandwidth C and surround signals, but 6dB coherent summing for the low frequency. The number should possibly be somewhere between 3dB and 6dB in practice. There was a thread here somewhere discussing that value from the other direction for main->subwoofer bass redirection. (The error accumulates the more speakers are involved in the redirection.)

I just googled for a while trying to find a definitive answer to this, but it is complicated (because Dolby is).
the thing is we need to look at what does the decoding. if we play a video on PC decoding happens in the video player (unless we output the sound digitaly, but this is not the usecase discussed here). So whether or not an AV reciever adds a boost during decoding is actualy irrelevant, because when we do a manual downmix we are not doing the decoding part. it has allready been done. this documents suggests that all the leveling is done by the decoder: https://developer.dolby.com/globalassets/professional/documents/dolby-metadata-guide.pdf
everything is done using parameters encoded in the signal, even the downmix. if we bypass the downmix by doing our own, the document actualy suggest that we would have to read out the downmix parameters, because the producer can choose between 3 levels for the center, for example. I think we can probably assume that the default parameter is beeing used though.
now the problem is that AC3 discards LFE during a downmix done by the decoder. there is no parameter for this, it is the default.
So what the hack do we do with LFE level during a manual downmix?
there is no metadata for the LFE boost at decoder stage. It seams it does exist though. So we must assume that it is a fixed boost. Unless somebody has a guideline from dolby for AVs that delgates this boost to them (outside of the decoding process).
So again, and I must repeat that this stuff is very confusing and I am not claiming to have the definitve answer here, but the leveling is done by the decoder. And doing a manual downmix doesn't bypass the decoder. that is at least my interpretation.
At the end our ears should indicate what the true answer is. 10dB is a hell of a boost. I have read a lot of people saying that it just don't sound right. I never tryed it out, but my level without the boost sounds right to me. I even might try it out so I can have my personal answer. but I can't imagine this sounding right, at least not in my setup. but I eventualy try it out

 

[KMO]

but my level without the boost sounds right to me.

One problem is what a tool or device defines as "without the boost". Often a "0" setting for LFE denotes the standard 10dB level difference. For example Denon AVRs have an LFE level setting ranging from -10dB to 0dB. "-10dB" means equal level, and "0dB" means standard 10dB-higher level.

10dB is a hell of a boost.

Well, it's not a "boost", is it? It's just the standard difference between LFE and other channels, which is why the Denon AVR calls it "0dB", and equal-level "-10dB". It would be just as true to say that playing it -10dB down is a hell of a cut.

PC tools seem to vary in how they regard the LFE and what they call "0dB". Every audio editing program I've seen misleadingly shows the LFE channel graphically the same size as the others, rather than 3x as tall, so all the LFE signals look very small, and a lot of tools don't handle that channel specially - it's just another channel, meaning you do have to make the 10dB adjustment yourself if transferring a signal between LFE and another channel. Some higher-level tools will do 10dB adjustment by default. And that is still configurable because there is content out there where the "LFE" channel is actually equal-level (eg from SACDs or early DTS CDs). Plenty of scope for confusion.

And another point is that the "boost" can't be done "inside" the decoder - you can't just boost a signal by 10dB, it'll clip. You need to have a common agreement that the LFE output data has a higher level, so 0dBFS for it means louder than 0dBFS on other channels. (What could be done is lowering all the other channels by 10dB to match, but that would be weird and defeat part of the point - to maintain good SNR on other channels).

A receiver can tell the Dolby decoder to output 5.1 or 2.0. When outputting 5.1, they mutually agree and understand that the .1 channel is 10dB louder than the rest, and the receiver handles it accordingly. When outputting 2.0, the Dolby decoder just throws the LFE away - any attempt to include it will likely cause clipping.

A receiver that wanted to try to include the LFE in a downmix could tell the Dolby decoder to output multichannel then do its own downmix, including special handling for providing headroom for the LFE (eg lowering the data digitally by 10dB and incorporating an equivalent analogue volume boost - costing you 10dB of noise headroom though).

Also, when processing non-Dolby data like multichannel PCM, the receiver is not seeing any metadata and has less idea what might be in the signal. Dolby is relying on production guidelines to assume that the LFE data is "redundant" so it chucks it. This might not be true for some random multichannel PCM source, so the receiver may well want to not chuck it.

 

[flaviowolff]

digitalfrost's mix calculation looks reasonable to me. Including the detail of assuming 3dB non-coherent summing for the full-bandwidth C and surround signals, but 6dB coherent summing for the low frequency. The number should possibly be somewhere between 3dB and 6dB in practice. There was a thread here somewhere discussing that value from the other direction for main->subwoofer bass redirection. (The error accumulates the more speakers are involved in the redirection.)

Hi! Does that mean that the level should vary from 1.58 (+4db) to 2.23 (+7db)?

 

[edechamps]

there is no metadata for the LFE boost at decoder stage. It seams it does exist though. So we must assume that it is a fixed boost.

I would assume it is, and will always be, +10 dB. That's because ITU-R BS.775, which is as close as it gets to a widely agreed standard for consumer multichannel audio, mandates +10 dB for LFE. I don't think anyone in their right mind would produce video content with a different LFE gain, as such content would almost surely be played back at the wrong level at the consumer.

(The only exception would be multichannel music such as SACD or DVD Audio, because these people had the absolutely brilliant idea to go against what the video industry was already doing, thereby guaranteeing their content will not play correctly on typical AV consumer systems. Genius!)

 

[KMO]

Hi! Does that mean that the level should vary from 1.58 (+4db) to 2.23 (+7db)?

Had to read back a bit. Yes, looking at digitalfrost's equations, you should get a reasonable result somewhere in that range. The 1.58 used there is possibly too low, but 2.23 is probably too high.

From the "why bass management makes my life tedious" thread you can see two real-life test examples in the 3dB-6dB range (corresponding to 4dB to 7dB in the equation) - L+R summing at +3.9dB and L+C summing at 5.9dB (which would then want +6.1dB or +4.1dB in the equation respectively).

If you did your own real-life test, you could probably come up with the optimal number for your system...

 

[KMO]

I would assume it is, and will always be, +10 dB. That's because ITU-R BS.775, which is as close as it gets to a widely agreed standard for consumer multichannel audio, mandates +10 dB for LFE. I don't think anyone in their right mind would produce video content with a different LFE gain, as such content would almost surely be played back at the wrong level at the consumer.

(The only exception would be multichannel music such as SACD or DVD Audio, because these people had the absolutely brilliant idea to go against what the video industry was already doing, thereby guaranteeing their content will not play correctly on typical AV consumer systems. Genius!)

Agree on the video content, and even music has now largely locked in on that, I believe. It's basically only SACD and early DTS-CDs with the equal-level. I'm under the impression all my DVD-Audios use standard LFE level, and Blu-ray audio discs do.

SACD seems to be consistently-enough equal-level (I think it is part of the standard) that "being DSD" is arguably interpretable as "low LFE level" metadata (and some receivers will remember the LFE level setting separately for PCM and DSD, achieving that effect).

When converting between DSD and PCM you probably want to do a LFE level conversion. Although that tends not to happen in normal consumer kit. Which in turn is actually one of the main annoyances about outputting DSD as PCM - the source has converted it with no level change, so the receiver can no longer see it was DSD, so the "low LFE" inference from that can't be done, and it doesn't automatically flip to the "LFE -10dB" setting.

But then my Denon AVR-4308 has a different bug with DSD over HDMI (which it didn't have over Denon Link) - it does seem to know that DSD should have lower LFE, so adjusts it down by 10dB automatically, but wrongly does it for the entire subwoofer output, including redirected bass. You'd have to set "LFE -10dB, SW +10dB" to get correct output. As adjusting manually adjusting the subwoofer to play SACD is more of a pain and more error-prone than manually adjusting LFE, I've settled for making the player output PCM.

 

[dasdoing]

I would assume it is, and will always be, +10 dB. That's because ITU-R BS.775, which is as close as it gets to a widely agreed standard for consumer multichannel audio, mandates +10 dB for LFE. I don't think anyone in their right mind would produce video content with a different LFE gain, as such content would almost surely be played back at the wrong level at the consumer.

(The only exception would be multichannel music such as SACD or DVD Audio, because these people had the absolutely brilliant idea to go against what the video industry was already doing, thereby guaranteeing their content will not play correctly on typical AV consumer systems. Genius!)

It is still not answering the question afaiui.
first, note that it's a recomendation for production. If you want the signal for LFE to be 10dB lower on the medium you have to add 10dB gain to it during production. that's the main point of the recomendation.
it doesn't explain though how you would manualy handle the LFE in a downmix as a consumer. Yes, at some point you need to add the gain. But my point was that this gain seams to be hardcoded into the decoder....and therefore if you add it again in manual downmix you added it twice; though KMO might have a point

 

[KMO]

It is still not answering the question afaiui.
first, note that it's a recomendation for production. If you want the signal for LFE to be 10dB lower on the medium you have to add 10dB gain to it during production. that's the main point of the recomendation.
it doesn't explain though how you would manualy handle the LFE in a downmix as a consumer. Yes, at some point you need to add the gain. But my point was that this gain seams to be hardcoded into the decoder....and therefore if you add it again in manual downmix you added it twice; though KMO might have a point

You're a bit confused. How would you "hardcode" gain into a "decoder"?

You're processing 16 or 24-bit digital signals. The digital "0dBFS" is the amplitude of maximum non-clipping sine wave.

The decoder (DD/DTS/TrueHD/FLAC/whatever) ultimately spits out PCM data. All the main channels work as they always do - they can have full-range signals up to 0dBFS. So how can the LFE channel be 10dB louder?

You can't do anything to the LFE data to make it louder - there's no headroom. The main channels are already using the full headroom. You can't have a +10dBFS signal without clipping.

The is the point of having the LFE channel in the first place - to reproduce signals so loud that you couldn't put them the main channels.

So the convention is that the LFE signal/data is reproduced 10dB louder. Its maximum 0dBFS point is 10dB louder than the other channels' 0dBFS. (peak 115dB SPL versus peak 105dB SPL).

So when you are downmixing, you have to take into account that 10dB level difference in the data.

Now, it would be possible for the decoder to output PCM data where all channels were equal, but only by making the main channels 10dB quieter, losing a couple of bits of signal-to-noise.

But if you ever see your decoded data having any signal peaks greater than -10dBFS in the main channels, you know it can't have done this.

All of this LFE stuff was originally done actually on magnetic recordings on 70mm film prints, for exactly the same reasons. They wanted louder bangs, but the only way to do that on the main channels without overloading the signal would be to record everything else 10dB lower on the print, increasing hiss. So they added the extra ".1" tracks that were played louder, but didn't have to worry about hiss due to their LPF.

TL;DR - if the "decoder" could do the gain adjustment and the data could just contain the level difference LFE wouldn't need to exist. They could have put the LFE signal in the main channels. It's entire raison d'etre is the different final gain levels, and the different interpretation of the data, be it analogue or digital.

 

[dasdoing]

You're a bit confused

I am pretty sure we all are.

How would you "hardcode" gain into a "decoder"?

LFE +10dB

how do you know there is no headroom for this in the encoded signal?
if you monitor it 10dB louder you will tend to bring it down electricly by 10dB. So the LFE is efectivly encoded quieter.
also note that the LFE is a LF extension. there is "enough" (by design) LF allready in the other channels. you don't need the extra headroom to make LF louder, since the sum will be louder anyways as you have an extra channel only for it

Anyways, I am definitly a little confused and this subject is low priority to me atm.

 

[KMO]

how do you know there is no headroom for this in the encoded signal?

We're kind of talking in circles here. There is headroom, but only if you interpret "0dBFS" (the maximum possible signal level) as being louder for the LFE channel.

If you regard a digital signal as giving you samples from -1.0 to +1.0, then a sine wave with peaks at -1.0 and +1.0 - the loudest you can go without clipping is "0dBFS".

A 0dBFS signal on the main channel corresponds to 105dB SPL. A 0dBFS signal on the LFE channel corresponds to 115dB SPL.

If you are doing are combining or transferring data between LFE and other channels that 10dB signal interpretation difference needs to be accounted for. A 105dB SPL signal (max main) would be from -0.3 to +0.3 on the LFE channel.

A 115dB SPL signal (max LFE) would be from -3 to +3 on a main channel, which would lead to clipping, so you can't do it, unless you apply an overall level adjustment lowering everything to give you headroom.

if you monitor it 10dB louder you will tend to bring it down electricly by 10dB. So the LFE is efectivly encoded quieter.

That much is true - 85dB SPL noise will be -20dBFS on the main channel, and -30dBFS on the LFE channel, because of that 10dB final gain offset.

But this is not part of the "encoding", so it's not something undone by the "decoder". It's part of the mix. It would have been that level in the PCM data on entry to the encoder, and will be that level in the PCM data on exit from the decoder. It's always the amplification stages of the monitoring, at both production end and playback end, that introduces the offset.

And the offset is totally independent of actual compression format. LPCM/DD/DTS/TrueHD/FLAC all store arbitrary PCM data, and in every case the LFE channel has the 10dB offset before encode and after decode. No encoder/decoder should change that.

also note that the LFE is a LF extension. there is "enough" (by design) LF allready in the other channels. you don't need the extra headroom to make LF louder, since the sum will be louder anyways as you have an extra channel only for it

That is somewhat true - the 10dB of extra headroom is roughly equivalent to sending the signal to 5 full-range speakers. But that's still extra headroom. You have 10 speakers' total worth of bass headroom, if sending bass to both main speakers and LFE.

 

[KMO]

As that was a lot of words, here's a visualisation from a real 5.1 disc - found this on Quadraphonic Quad - it's a from Queen's A Night at the Opera DVD-Audio.

That is the L, R, C, LFE ,Ls, Rs tracks in that order. The centre is quite lightly used (it's nearly a 4.1 mix), and as you can see the LFE track is tiny. But that's because, as discussed above, mixing tools just process the LFE track like any other. Each track has the [-1.0, 1.0] range, and they're shown the same height. But the LFE track will be played 10dB louder, so when it's displayed in the same width, it looks small.

Arguably the waveform editor would do well to have an option to display the LFE track 3 times (10dB) taller, then its waveform amplitude would look more like the others, and you could say that the tripled space was the extra headroom.

One final way of looking at it - the LFE is measured in different units. It's as if it's in inches rather than centimetres, so its "1.0" is a lot bigger...

 

[tuckers]

To head back to the posters original thread, I also have an Nvidia Shield Pro and send the output to a 2 channel DAC and I have speakers that go to 25hz. I know your problem! This only really became an issue for me with the Shield upgrade to Android 11 which broke the ability of some media apps to connect with network attached storage (this bug is actually the fault of the apps not being updated to Android 11 specs on file permissions). My app of choice was MrMC, which had the option to downmix the LFE channel to the 2 channel mix. I know of no other working app that offers this. Unfortunately MrMC hasn't been updated in almost 2 years on Android, and the NAS issue won't be fixed, so I had to find another solution. I have spent countless hours researching, finding only a partial solution.

First off, I have no problems with any of the streaming platforms, Netflix HBO etc., all mix in the LFE track or have a full stereo track embedded in their content already. The problem is in 3rd party video apps like Kodi, etc. Most do not send the LFE track to a 2 channel audio system. I don't know if this is the fault of the app or the Shields 2 channel downmixing, but I think it is the apps. The one app I've found that does pass the LFE track is VLC. So that is what I am using. It only works when I have the Shields audio settings set to downmix to 2 channel and Dolby Digital enabled. No other modes work. I use USB audio out for this. I have also tried 5 different HDMI Audio converter boxes with Toslink out, and I have had mixed results, VLC in that configuration is the only way it passes LFE.

First off was finding a test track that is in Dolby digital format and had a LFE test in it. I found one after much searching. About 3 minutes into the track is where the LFE test begins. You can download this and put on your media streamer locally. Try this track using different media apps, you will get different results with each. Most apps do not send the LFE channel when the system is set to 2 channel. This issue is compounded by the fact that many surround mixes (at least the mixes that come from the video transcodings of most MKV files etc.) don't really follow the best practices about the L+R channels, and cut off the frequencies under about 80hz and send the lower frequencies to the LFE instead. Hence you don't get any frequencies below about 80hz when you have your system setup in 2 channel.

So for me I use VLC and set my system up to use Dolby digital in the sound settings when I want to play move and TV show files. Then I have to reset it to non Dolby Digital when I'm done, because I don't like what it does with other apps.

I understand the MrMC is still actively supported as an Apple TV app, so I might bite the bullet and try an Apple TV, but this is a hard ask, since I strongly dislike the Apple Ecosystem.

Good luck!

 

[dasdoing]

So I said I would test out a boost. Unfortunatly I don't watch movies very often atm. Can't realy find intrest in majority what is beeing produced atm.
I tested with the latest 007, tough. that scene where he is with his girlfiriend and they end up beeing shot at. one of the "bag guys" comes closer and starts to shot rounds at the shielded window.
fist thing I noted is that the LFE that comes out of VLC hits 0dBFS at one point.
now I have the headroom to add extra 10dB (efectivly bringing the mains down).
those shots at the window sound stupidly unrealistic with the boost. the bass is almost as loud as the "bang".
now this proves nothing, it could be that it was intended to sound stupid like this, OR (and that's my guess) VLC actualy boosts LFE in the decoding process.
whatever the real answer is, I quickly decided to watch the movie without the boost. will test more though, once I find a movie I am intrested in

 

[KMO]

now I have the headroom to add extra 10dB (efectivly bringing the mains down).

I'm not clear what your original setup was - it may be that you don't need to add anything because the boost has always been in there somewhere in your setup.

The standard behaviour is to have LFE higher than mains, so you shouldn't need to set any control to "+10dB" anywhere in most normal apps that understand the concept of LFE. Removing the LFE boost is often indicated as "-10dB" versus the standard "0dB".

The +10dB being talked about above was when someone was performing their own manipulation in a generic equation - they had to remember to include it in their maths.

fist thing I noted is that the LFE that comes out of VLC hits 0dBFS at one point.
... OR (and that's my guess) VLC actualy boosts LFE in the decoding process.

That's certainly loud then, but you have to ask yourself, if VLC was actually boosting before the point you monitored it, what would it have done with a real 0dBFS LFE signal? Try to make it +10dBFS and clip? I hope it's not boosting with potential clipping, or you'll get something unpleasant at some point.

 

[flaviowolff]

To head back to the posters original thread, I also have an Nvidia Shield Pro and send the output to a 2 channel DAC and I have speakers that go to 25hz. I know your problem! This only really became an issue for me with the Shield upgrade to Android 11 which broke the ability of some media apps to connect with network attached storage (this bug is actually the fault of the apps not being updated to Android 11 specs on file permissions). My app of choice was MrMC, which had the option to downmix the LFE channel to the 2 channel mix. I know of no other working app that offers this. Unfortunately MrMC hasn't been updated in almost 2 years on Android, and the NAS issue won't be fixed, so I had to find another solution. I have spent countless hours researching, finding only a partial solution.

First off, I have no problems with any of the streaming platforms, Netflix HBO etc., all mix in the LFE track or have a full stereo track embedded in their content already. The problem is in 3rd party video apps like Kodi, etc. Most do not send the LFE track to a 2 channel audio system. I don't know if this is the fault of the app or the Shields 2 channel downmixing, but I think it is the apps. The one app I've found that does pass the LFE track is VLC. So that is what I am using. It only works when I have the Shields audio settings set to downmix to 2 channel and Dolby Digital enabled. No other modes work. I use USB audio out for this. I have also tried 5 different HDMI Audio converter boxes with Toslink out, and I have had mixed results, VLC in that configuration is the only way it passes LFE.

First off was finding a test track that is in Dolby digital format and had a LFE test in it. I found one after much searching. About 3 minutes into the track is where the LFE test begins. You can download this and put on your media streamer locally. Try this track using different media apps, you will get different results with each. Most apps do not send the LFE channel when the system is set to 2 channel. This issue is compounded by the fact that many surround mixes (at least the mixes that come from the video transcodings of most MKV files etc.) don't really follow the best practices about the L+R channels, and cut off the frequencies under about 80hz and send the lower frequencies to the LFE instead. Hence you don't get any frequencies below about 80hz when you have your system setup in 2 channel.

So for me I use VLC and set my system up to use Dolby digital in the sound settings when I want to play move and TV show files. Then I have to reset it to non Dolby Digital when I'm done, because I don't like what it does with other apps.

I understand the MrMC is still actively supported as an Apple TV app, so I might bite the bullet and try an Apple TV, but this is a hard ask, since I strongly dislike the Apple Ecosystem.

Good luck!

This is one of the many reasons why a HTPC is the best solution.

 

[dasdoing]

I'm not clear what your original setup was

VLC into EQ-APO where I do manual downmix like this

Copy: L=L+-3.0dB*C+-3.0dB*LFE+-3.0dB*RL R=R+-3.0dB*C+-3.0dB*LFE+-3.0dB*RR C=0 LFE=0 RL=0 RR=0 SL=0 SR=0

when testing with the extra boost I change "+-3.0dB*LFE" to "++2.0dB*LFE"

what would it have done with a real 0dBFS LFE signal? Try to make it +10dBFS and clip?

probably. good thing windows has a limiter lol
and actualy 0dBFS is a clip most of the time. it was just a fraction of a second though iirc when a car crashed into their car

 

[KMO]

So the first question is whether Equalizer-APO already has a +10dB adjustment built-in for the LFE channel when you give it a mix equation. If it is aware of what LFE is, and that it is a louder signal then it may well do.

That would make more sense than a belief that VLC would be doing an adjustment.

Edit: A bit of research doesn't show any evidence that there is a built-in 10dB adjustment. Here I see someone doing the opposite-direction 10dB adjustment manually in a bass-management filter:

Eval: crossfreq=120
Copy: BASS=L+R+C+SL+SR
Channel: L R C SL SR
Filter: ON HP Fc `crossfreq` Hz
Filter: ON HP Fc `crossfreq` Hz
Channel: BASS
Filter: ON LP Fc `crossfreq` Hz
Filter: ON LP Fc `crossfreq` Hz
Copy: LFE=LFE+-10dB*BASS

 

[edechamps]

So the first question is whether Equalizer-APO already has a +10dB adjustment built-in for the LFE channel when you give it a mix equation. If it is aware of what LFE is, and that it is a louder signal then it may well do.

No it does not. Equalizer APO treats LFE just like any other channel.

My understanding is that, in general, whatever component is doing the bass management is also responsible for handling the +10dB LFE gain. So, for example, if the AV receiver is doing the bass management, it will apply the +10 dB gain. When manually implementing bass management in Equalizer APO, you are on the hook for making sure the +10 dB gain is applied. To me, it logically follows that whatever is doing the downmixing to stereo is responsible for applying the LFE gain as well, as otherwise it won't happen anywhere else. Therefore, if you're doing manual downmixing in Equalizer APO, you are on the hook for manually applying the +10 dB gain there, and applying the necessary attenuation to avoid clipping.

 

[KMO]

What does "apply the necessary attenuation" entail?

In that bass management snippet above, someone is just adding channels willy-nilly. Is it using a "wide" or floating-point intermediate internally, so that's not a problem, you just need to make sure the final output result to the system doesn't clip?

(The convention I usually see in home theatre gear is that sub channels including bass management usually use a 15dB offset for the sub, so that last step would normally be

SUB=-5dB*LFE + -15dB * BASS

to help avoid clipping.)