Hugo M Scaler by Chord Electronics - The Official Thread
post-14888700
Post #6,496 of 11,688

Christer

1000+ Head-Fier
Joined
Nov 23, 2015
Messages
1,276
Reaction score
998
Joined
Nov 23, 2015
Posts
1,276
Likes
998
     Share This Post       
  • Like
Reactions: 514077 and miketlse
post-14888748
Post #6,497 of 11,688

analogmusic

100+ Head-Fier
Joined
Apr 15, 2015
Messages
452
Reaction score
197
Joined
Apr 15, 2015
Posts
452
Likes
197
My limited understanding is even if we don't directly hear those high frequencies, the hetrodining between them can add a little colour to the timbre of a given instrument. Which may be one reason why some people find 44.1k digital a bit dry or lacking exactly because of the removal of partials above 20k. Wazzzzup made some through-the-nose snark remark last time I brought this up. But, I don't bother with bar-room smugness anyway. But, until someone refutes my naïve thoughts with some actual facts, I'm stickin' to what I think is the case.
Hell, we even talked about this back in engeneering school back in the '80s. The digital experts said things that add up to: Don't worry yer perdy little head about it; 20k is all you need. On the other hand, I talked to some actual sound engineers who felt that pushing the cutoff from 22.05 up to about 50K would have been better in the beginning. So we'd have had to live with perhaps 8" or 10" disks. Some whine about missing the 12" cover of an LP with all the pictures and album credits.
From what I can hear it’s more to do with transients than ultrasonics ....

Try to play a 20khz test tone and report back if you can hear it

Pushing the sampling rate up has the effect of reducing the transient error....

In the morning I was listening to Hans Zimmers live in Prague on my mojo and it sounded gloriously musical ...
 
Last edited:
     Share This Post       
post-14888750
Post #6,498 of 11,688

Christer

1000+ Head-Fier
Joined
Nov 23, 2015
Messages
1,276
Reaction score
998
Joined
Nov 23, 2015
Posts
1,276
Likes
998
My limited understanding is even if we don't directly hear those high frequencies, the hetrodining between them can add a little colour to the timbre of a given instrument. Which may be one reason why some people find 44.1k digital a bit dry or lacking exactly because of the removal of partials above 20k. Wazzzzup made some through-the-nose snark remark last time I brought this up. But, I don't bother with bar-room smugness anyway. But, until someone refutes my naïve thoughts with some actual facts, I'm stickin' to what I think is the case.
Hell, we even talked about this back in engeneering school back in the '80s. The digital experts said things that add up to: Don't worry yer perdy little head about it; 20k is all you need. On the other hand, I talked to some actual sound engineers who felt that pushing the cutoff from 22.05 up to about 50K would have been better in the beginning. So we'd have had to live with perhaps 8" or 10" disks. Some whine about missing the 12" cover of an LP with all the pictures and album credits.

Interesting thoughts you raise there.

I have also discussed this and other SQ issues with quite a few engineers at actual recording sessions of large scale classical symphonic music.
And I have received answers with completely differing opinions.

No names, but one big name in the classical recording business said that "what really matters is 24 bits, but a sampling rate of 44.1 is enough."

Another one said "I can live with 20 bits but a sampling rate of at least 192 khz is vital".

And others say that DSD 256 is the only way to go.

One of them particularly mentioning the very low level spatial cues one hears with acoustic music in a real hall as something DSD 256 captures better than the current rivalling PCM format DXD.

With so much contradictory information from the people actually working with and making recordings I hope I am a bit excused if I am a bit confused ,to say the least.

But I have to say that now with the M Scaler and 24 bits the 44.1 khz sampling rate which I could earlier quite easily hear that it sounded less realistic than 88..2 or 96 and above is NOT a problem any longer for me.

Strings and particularly massed strings which tended to be THE give away with 24/44.1 recordings before M Scaler now tend to sound as good as higher rates to me.

But I still hear a difference between 16 bits and 24 bits.

So when I say HI RES is needed, I now refer to 24 bits over 16 bits
.
M Scaler seems sort out the sampling rate issue for me and my ears/brain.
But Qutest on its own does not!

But what if 32 or 64 bits as recording formats would also make a difference?
And why will Davina record at 768khz?

And in spite of Rob's take on DSD,there are still recording labels and engineers that swear by DSD 256 and claim nothing gets closer to the actual sound of acoustic instruments in a hall than DSD 256?
Some will vehemently defend their stand.

All I can say is that to me DSD can also sound good, even very good if it hasn't been edited.
Both the way Rob does it with M Scaler and natively.
Maybe we are each of us sensitive to slightly different aspects of SQ?

According to one person whose opinions I value,and who has a SOTA system, for Rob Watts depth matters above everything else obviously?
But as I said in an earlier post here at most live concerts I attend, and I attend lots of them, I hear more width than depth.
And to me timbre and tonality of acoustic instruments and the human voice are absolutely paramount.
Sibilance on female voices is not anything I have EVER heard live but with some digital it can be quite a problem.
Anyway ,since I opted out of Canjam this year for live music,I hope to see Rob next year at Canjam or even better, before Canjam Singapore with a DAVINA and some DX amps hopefully with headphone ports included!

Cheers CC
 
Last edited:
     Share This Post       
post-14888764
Post #6,499 of 11,688

Christer

1000+ Head-Fier
Joined
Nov 23, 2015
Messages
1,276
Reaction score
998
Joined
Nov 23, 2015
Posts
1,276
Likes
998
From what I can hear it’s more to do with transients than ultrasonics ....

Try to play a 20khz test tone and report back if you can hear it

Pushing the sampling rate up has the effect of reducing the transient error....

In the morning I was listening to Hans Zimmers live in Prague on my mojo and it sounded gloriously musical ...
Hmm, more to do? Or ALL to do?
Cheers CC
 
     Share This Post       
post-14888784
Post #6,501 of 11,688

Rob Watts

Member of the Trade: Chord Electronics
Joined
Apr 1, 2014
Messages
2,406
Reaction score
6,536
Joined
Apr 1, 2014
Posts
2,406
Likes
6,536
@Rob Watts, do you sit back and watch us poorly informed mortals discussing your chosen profession to death. Only for you to jump in at the 11th hour and squash us.

Keep it up its very entertaining it really is.
No I don't, I very much enjoy reading posts; but what angers me is the industry as a whole jumping onto the ringing is bad, or brick-wall filters are evil meme; and I can't figure out what the explanation for this fallacy is - either the manufactures are just plain ignorant of sampling theory, or that the whole ringing is bad fallacy is just a simple message to con audiophiles into buying something; frankly I don't know which explanation is worse, marketing or technical ignorance. I suspect the drivers for this is actually both.

Actually I'm still hesitating to say ultrasonics do absolutely nothing, since there's some theoretical scenarios where they could come into play, despite being inaudible as such. On the other hand I tend to believe Rob in this respect. So definitely slightly undicided yet. However, even if they're audible under certain circumstances, they won't play a decisive role, that's what I think.
Yes ultrasonic overtones could become audible in the presence of some non-linearity, as the overtones with the non-linearity would create audible (within the audio bandwidth) intermodulation products. The air itself is non-linear, but the mic would actually pick-up these intermod products anyway. The other area of non-linearity is the ear itself. But - I have spent my whole career trying to eliminate intermodulation (noise floor modulation etc) and I can't see how adding more intermodulation would be a good thing - my view is that this is likely to be a tiny effect, and one where actually bandwidth limiting would help sound quality. Intermodulation is always a bad thing in my experience.

Having said all of the above, we will know for sure when I actually bandwidth limit 768k files without decimating it, maintaining 768k.

Rob, in this case John Atkinson used an impulse response as a criterion for showing the characteristic of the ringing, which indeed looks like an illegal signal. On the other hand, if he had used a square-wave response for the same purpose, I bet the result would look very similar if not identical. At least that's what usual measurements with both measuring criteria illustrate. And a square wave is definitely a legal signal.

[Edit:] Of course I'm talking of a bandwidth-limited square-wave signal – doing otherwise would be downright criminal.

It's always been my belief that any low-pass filter that's not infinitely sharp will be «excited» also by frequencies minimally below its own frequency, so in this context the ringing would be unavoidable in any case. Or where's my fallacy?
Using an impulse test is a perfectly valid way of defining a filters performance, as for a FIR filter it will return the coefficients directly. What is wrong is using an illegal signal to infer SQ changes...

You also need to draw a distinction between reconstruction or interpolation filters used with bandwidth limited signals too; a filter that had pre-ringing for EQ (that is ringing within the audio bandwidth) would definitely sound strange; I would never use FIR filters for audio EQ purposes, because of the pre-ringing. But that's a very different scenario to the reconstruction case - for which if you want to perfectly reconstruct you must use a Sinc function filter, which infinitely rings with illegal impulses. But won't ring at all (both pre and post ringing) with bandwidth limited impulses; it will just return the original un-sampled impulse perfectly.

Let’s say it’s true that ultrasonic frequencies generated by an instrument color the audible spectrum. If so, the mic capturing that audible spectrum will pick that up and record it in the audible spectrum, to be played back as originally heard. Reproducing the ultrasonic component. Isn’t necessary.
Yes agreed - but we won't know for sure until we do listening tests to confirm that ear non-linearity isn't important too.

From what I can hear it’s more to do with transients than ultrasonics ....

Try to play a 20khz test tone and report back if you can it

Pushing the sampling rate up has the effect of reducing the transient error....
Absolutely - the discussion about ultrasonics audibility and ringing is akin to talking about a hair on the top of an elephants head - whilst ignoring the elephant in the room. The elephant in the room is the failure of conventional interpolation filters to accurately reconstruct the timing of transients accurately. And it's a psychoacoustic fact that transients are the most important cue that the brain uses to make sense of the sound scene; when transients are constantly shifting backwards and forwards it destroys the brain's ability to decode timbre, pitch, starting and stopping of notes and separate instruments out.
 
post-14889046
Post #6,502 of 11,688

delirium

100+ Head-Fier
Joined
Mar 20, 2016
Messages
327
Reaction score
79
Joined
Mar 20, 2016
Posts
327
Likes
79
I recently moved from the supplied optical cable in the M Scaler box to a Qed reference digital optical cable and the difference is very noticeable

Maybe it is source dependent but i would definitely try a glass cable if you can

Just my thoughts

Deano
Thanks
 
     Share This Post       
  • Like
Reactions: Deano1974
post-14889344
Post #6,503 of 11,688

514077

Headphoneus Supremus
Joined
Mar 9, 2013
Messages
3,711
Reaction score
1,557
Joined
Mar 9, 2013
Posts
3,711
Likes
1,557
From what I can hear it’s more to do with transients than ultrasonics ....

Try to play a 20khz test tone and report back if you can hear it

Pushing the sampling rate up has the effect of reducing the transient error....

In the morning I was listening to Hans Zimmers live in Prague on my mojo and it sounded gloriously musical ...
To loosely quote yourself,"you missed the point'. Anyone knows you can't hear a direct 20k note, especially at my age. But, what of the subsonic beats produced by a 20k and a 19k tone, for example? But I'm not interested in getting into a pi...posting contest with you.
@Jarnop said: Let’s say it’s true that ultrasonic frequencies generated by an instrument color the audible spectrum. If so, the mic capturing that audible spectrum will pick that up and record it in the audible spectrum, to be played back as originally heard. Reproducing the ultrasonic component. Isn’t necessary." You know, that's an interesting point.
 
Last edited by a moderator:
     Share This Post       
post-14889364
Post #6,504 of 11,688

analogmusic

100+ Head-Fier
Joined
Apr 15, 2015
Messages
452
Reaction score
197
Joined
Apr 15, 2015
Posts
452
Likes
197
Ho hum...... not sure if you truly grasped and understand what Rob Watts just clarified...

Nothing at even 19 kHz can be heard... try it.

I’m not the one talking about the hair on the elephants head while missing the whole elephant

Some of Rob Watts explanations clearly lost on some people here.

I will concede there’s reverbation of high frequencies but that’s down to the recording venue and position of mikes

Nothing to do with a Chord Dac.
 
Last edited:
     Share This Post       
post-14889497
Post #6,505 of 11,688

Amberlamps

Formerly known as Phuca
Joined
Jan 22, 2017
Messages
2,562
Reaction score
1,417
Location
Amamufukinstarboy
Joined
Jan 22, 2017
Location
Amamufukinstarboy
Posts
2,562
Likes
1,417
Ringing can't be heard for two reasons:

1. Nobody can hear 22.05 kHz.

2. A Dirac impulse that is used to give you an impulse response that is used to show the ringing is an illegal signal from sampling theory point of view, as it contains equal energy at FS/2 (22.05 kHz) when sampling requires bandwidth limiting, which means there is no energy at 22.05 kHz. No energy on the input means the filter will in practice not ring at all.

This basic and gross misunderstanding of simple digital sampling theory has been adopted by the whole audio business; indeed, the M scaler with it's huge tap length WTA filter, with a bandwidth limited analogue signal will actually reconstruct the original analogue signal to a better than 16 bit accuracy; with no added ringing at all.

It is pretty simple really.

No I don't, I very much enjoy reading posts; but what angers me is the industry as a whole jumping onto the ringing is bad, or brick-wall filters are evil meme; and I can't figure out what the explanation for this fallacy is - either the manufactures are just plain ignorant of sampling theory, or that the whole ringing is bad fallacy is just a simple message to con audiophiles into buying something; frankly I don't know which explanation is worse, marketing or technical ignorance. I suspect the drivers for this is actually both.



Yes ultrasonic overtones could become audible in the presence of some non-linearity, as the overtones with the non-linearity would create audible (within the audio bandwidth) intermodulation products. The air itself is non-linear, but the mic would actually pick-up these intermod products anyway. The other area of non-linearity is the ear itself. But - I have spent my whole career trying to eliminate intermodulation (noise floor modulation etc) and I can't see how adding more intermodulation would be a good thing - my view is that this is likely to be a tiny effect, and one where actually bandwidth limiting would help sound quality. Intermodulation is always a bad thing in my experience.

Having said all of the above, we will know for sure when I actually bandwidth limit 768k files without decimating it, maintaining 768k.



Using an impulse test is a perfectly valid way of defining a filters performance, as for a FIR filter it will return the coefficients directly. What is wrong is using an illegal signal to infer SQ changes...

You also need to draw a distinction between reconstruction or interpolation filters used with bandwidth limited signals too; a filter that had pre-ringing for EQ (that is ringing within the audio bandwidth) would definitely sound strange; I would never use FIR filters for audio EQ purposes, because of the pre-ringing. But that's a very different scenario to the reconstruction case - for which if you want to perfectly reconstruct you must use a Sinc function filter, which infinitely rings with illegal impulses. But won't ring at all (both pre and post ringing) with bandwidth limited impulses; it will just return the original un-sampled impulse perfectly.



Yes agreed - but we won't know for sure until we do listening tests to confirm that ear non-linearity isn't important too.



Absolutely - the discussion about ultrasonics audibility and ringing is akin to talking about a hair on the top of an elephants head - whilst ignoring the elephant in the room. The elephant in the room is the failure of conventional interpolation filters to accurately reconstruct the timing of transients accurately. And it's a psychoacoustic fact that transients are the most important cue that the brain uses to make sense of the sound scene; when transients are constantly shifting backwards and forwards it destroys the brain's ability to decode timbre, pitch, starting and stopping of notes and separate instruments out.

The only thing I have to say on the subject is;

At it's worst, my nerves could physically feel audio from the tv or my speakers in the living room.

I schit you not, all I had to do was stick my feet up and aim them at the tv and deep bassy type noises would register and make my feet feel like they were standing on top of a speaker, like that dude in the movie, "it's all gone pete tong". My feet could feel the beat from the other side of the room and not just bass, the soles of my feet would feel like it was rippling and I could feel them change with different frequencies, it was nuts. Tv was slightly louder than normal.

Then it started happening in my hands, but not so bad, and now I'm left with a permanently gimped left thumb, if I touch the underside of my left thumb, I feel it on my left big toe, it's like somebody tickling my big toe, it's weird as schit, but, on the bright side, if I ever have an itchy toe, I just need to rub my thumb. :)

Im positive that if a bat was in the room with me then and it was in the dark, I'm sure my feet radar would of detected it's ultrasonic squeaks.


As for the real topic, I think that although 20khz is max for the best human ears on the planet ( young kids only, and even then it would probably be a low percentage of them that could hear 20khz ) to phsyically hear.

However, I do think that it is possible that ultrasonic sound has some influence, the same with the other end of the spectrum, infrasonic. Nobody can hear better than 20khz, but I do think infra and ultrasonic frequencies have some influence on us/brain, even though we physically cant hear them.

I bet that if someone were to do eeg's, ct scan's and mri scans with contrast and play normal music, and then play the same tracks but this time ultrasonic and infrasonic tones added, or even just scan them with no music, and then send both those tones down the headset to them without telling them they were doing it. I'm sure the results would show activity in certain parts of the brain that wasn't showing activity before both frequencies were played.

I would put money on that being true, as our brain is highly under utilised and alot of things happen to us subconciously and we just don't know about it.

Thats what I think regarding the topic, and I thought I would share a funny anecdote.
 
Last edited:
post-14889557
Post #6,506 of 11,688

JaZZ

Headphoneus Supremus
Joined
May 9, 2002
Messages
9,650
Reaction score
1,617
Location
Zürich, Switzerland
Joined
May 9, 2002
Location
Zürich, Switzerland
Posts
9,650
Likes
1,617
Age
70
You also need to draw a distinction between reconstruction or interpolation filters used with bandwidth limited signals too; a filter that had pre-ringing for EQ (that is ringing within the audio bandwidth) would definitely sound strange; I would never use FIR filters for audio EQ purposes, because of the pre-ringing. But that's a very different scenario to the reconstruction case - for which if you want to perfectly reconstruct you must use a Sinc function filter, which infinitely rings with illegal impulses. But won't ring at all (both pre and post ringing) with bandwidth limited impulses; it will just return the original un-sampled impulse perfectly.
Since I wasn't able to find an example with a perfect sinc-function filter, a typical 1 kHz square-wave response from an unnamed CD player will have to do it:



The typical Gibbs phenomenon – another name for the «ringing» – apparently can't be avoided with a square-wave signal from a bandwidth-limiting sampling frequency. Now it's obvious that above signal wasn't recorded from a bandwidth-limited original. Do you have the response to a bandwidth-limited (1 kHz) square-wave signal at your disposal that you can show us (e.g. from Hugo₂, TT₂ and/or Dave, with and/or without M Scaler)?
 
Last edited:
     Share This Post       
  • Like
Reactions: Christer
post-14889600
Post #6,507 of 11,688

tekkster

500+ Head-Fier
Joined
Sep 9, 2016
Messages
652
Reaction score
502
Joined
Sep 9, 2016
Posts
652
Likes
502
Very interesting stuff. So glad to see Rob's responses. Quite educational to read everyone else's responses too.

I've been rather overly focused these days on trying to figure out the slew rate of various DAC/amps, rather than the ringing phenomenon, and have been thinking about picking up an education grade or consumer grade oscillator for a few hundred dollars. A lot of the ones in the $100-$500 range include square wave generators, though I was imagining using a square wave sound generator from my iMac, through the H2, Mojo, ADI-2, BL, etc. and then use an adaptor to convert 6.3 / 3.5 / XLR -> BNC to connect back into the oscillator to see results.

All hobby level interest, as the audioscience website almost never includes slew rate in their tests.

Fun stuff, great reading, happy to be in this thread.
 
     Share This Post       
post-14890608
Post #6,508 of 11,688

rkt31

Headphoneus Supremus
Joined
Oct 14, 2014
Messages
1,569
Reaction score
454
Joined
Oct 14, 2014
Posts
1,569
Likes
454
what filter on H2/tt2/qutex with m scaler are people using, white or green and why ?
 
     Share This Post       
post-14890644
Post #6,509 of 11,688

Rob Watts

Member of the Trade: Chord Electronics
Joined
Apr 1, 2014
Messages
2,406
Reaction score
6,536
Joined
Apr 1, 2014
Posts
2,406
Likes
6,536
Since I wasn't able to find an example with a perfect sinc-function filter, a typical 1 kHz square-wave response from an unnamed CD player will have to do it:



The typical Gibbs phenomenon – another name for the «ringing» – apparently can't be avoided with a square-wave signal from a bandwidth-limiting sampling frequency. Now it's obvious that above signal wasn't recorded from a bandwidth-limited original. Do you have the response to a bandwidth-limited (1 kHz) square-wave signal at your disposal that you can show us (e.g. from Hugo₂, TT₂ and/or Dave, with and/or without M Scaler)?
I am assuming you mean that by bandwidth limited its set to FS/2? So CD would be flat at 20 kHz, and 0 at 22.05 kHz?

I have designed decimation filters - both IIR and FIR - for the Davina project - but although I don't have 1kHz responses to a 768 kHz derived 1kHz square wave, I can tell you what it would look like. The FIR response would look exactly like your image - pre ringing and post ringing - but the IIR filter would only have post ringing.

So they would look very different. The odd thing is that the odd order harmonics of the square wave would be exactly the same level - the reason that the square wave would look very different (only ringing after the edge) is because the phase of each harmonic is different, and the changes in phase would give the different waveforms. Symmetric FIR filters are phase linear (that is the delay is fixed irrespective of frequency) but IIR filters are not phase linear (delay changes with frequency).

Sampling theory has nothing to say on how the bandwidth limiting occurs - only that at FS/2 and above the level must be zero. My filters meet this requirement with 250 dB rejection - but they may end up sounding very different of course. Hence the listening tests I have planned, and why I have designed different filters.

As too the M scaler - with a regular square wave input set to 1 kHz the effect of the M scaler would be the same. That's because a square wave is not a transient, it's a regular signal with fixed harmonics at 21 kHz max (with BW limiting at 22 kHz say). Dave's response at 21k is identical to the M scaler's response at 21k. You need transients, which will have energy up to 22.05k to be able to see the change.

Edit: as a postscript, the FIR filter is actually more accurate - as the FFT of the square wave harmonics are exactly the same as the original non-bandwidth limited square wave - so the audible part of the square wave is exactly identical - even though the pre-ringing looks odd. Its the IIR filter response that actually changes the harmonics, or changes the audio bandwidth signal, even though it superficially looks better (no pre-ringing). My guess is the FIR decimation will sound better (i.e. closer to the original non bandwidth limited) as the harmonics within the audio part is completely unchanged.
 
Last edited:
     Share This Post       
post-14890692
Post #6,510 of 11,688

Christer

1000+ Head-Fier
Joined
Nov 23, 2015
Messages
1,276
Reaction score
998
Joined
Nov 23, 2015
Posts
1,276
Likes
998
Very interesting stuff. So glad to see Rob's responses. Quite educational to read everyone else's responses too.

I've been rather overly focused these days on trying to figure out the slew rate of various DAC/amps, rather than the ringing phenomenon, and have been thinking about picking up an education grade or consumer grade oscillator for a few hundred dollars. A lot of the ones in the $100-$500 range include square wave generators, though I was imagining using a square wave sound generator from my iMac, through the H2, Mojo, ADI-2, BL, etc. and then use an adaptor to convert 6.3 / 3.5 / XLR -> BNC to connect back into the oscillator to see results.

All hobby level interest, as the audioscience website almost never includes slew rate in their tests.

Fun stuff, great reading, happy to be in this thread.

Hmm, let me once again reveal my ignorance of tech terms. I have seen the term slew rate before,mainly in specs for amps I think?
But what exactly is it?

And why would it matter with dacs?

And while I am at it and in view of what Rob just replied regarding ringing the Gibbs phenomenon, and similar wasn't one of the main claims for DSD that it does not "ring" as PCM?

I vaguely remember seeing similar impulse pics comparing PCM and DSD where PCM showed ringing and the DSD did not. In my ignorance I did of course think that DSD would have better and cleaner transients than PCM with those shots as indications and evidence of that?
And one of the engineers I know used to claim transients with DSD as being closer to how he heard things live than with pcm, while another one claimed the opposite!
But if I now understand Rob correctly that is not true at all, the way he filters pcm?

And also,no matter how high you go with DSD you still have the fundamental limitations of the format?

So the new DAC from the German company T+A standing for "theory and practice" in German, which now claims to do both DSD 1024 and 3 billion computations per sec in their latest flagship, would still suffer the same basic limitations as any DSD dac?

But would not actually sound better or more resolving than say a Qutest on its own in purely technical terms if I understand Rob correctly?

But they still charge 26 K USD for it obviously? The Qutest sells for under 2k USD doesn't it?

And there is another native 1 bit DSD DAC just reviewed over at CA/AS that apparently costs 45K USD!

And I used to think Chord products were overpriced!

Maybe I will have to re-access my take on pricing a bit, in view of what I keep hearing via HMS and what Rob claims and what HMS costs?
Cheers CC.
 
Last edited:
     Share This Post       
  • Like
Reactions: 514077

Users Who Are Viewing This Thread (Users: 3, Guests: 4)

Top