Petition to stop FM being switched off

On Sun, 5 Jul 2009 16:24:19 +0100, Paul Martin
wrote:

In article ,
Stephen wrote:
On Sat, 4 Jul 2009 16:39:24 +0100, Paul Martin
wrote:

Modified delta modulation only gets you so far. The best "lossless"
compression system gets you down to about 50% of the original (raw)
bandwidth. ADPCM is about the same. SACD uses a delta encoding.

The worry here is that you are talking about how to do this, but
ignoring standards. The old joke and standards is it is much easier to
invent your own than to go find someone who already solved the
issue...

enhanced APTx seems to be fairly popular with broadcasters - it gives
4:1 compression, but it is proprietary. The license owner is APT is
Belfast, but the scheme is in many manufacturers purpose built codecs.
http://www.aptx.com/content.asp?page=104&site=2

It does a good job for stereo 15 Khz at 256 / 384 Kbps data rate and
really well for 22 KHz at 576 Kbps (but i am a datacomms person - no
doubt the audiophiles have a different opinion).

There's nothing in that which says it's a "lossless" codec, like
Monkey's Audio, FLAC or MLP. Depending on the material, you *can* get
4:1 compression with lossless codecs, but it's not consistent.

i didnt say it was lossless.

22kHz at 576kbps is broadly equivalent to NICAM's 32kHz at 728kbps.

Just to put your figures in perspective, 15kHz at 16 bit stereo is
480kbps UNcompressed, and 22kHz is 704kbps UNcompressed. You're quoting
compression ratios of about 18%.

Not sure about your maths, but i think you are ignoring the Nyquist
sampling issue.

15 Khz needs to be sampled at 32 K samples / sec.
With 16 bit samples and 2 channels, that is 1024 Kbps.

4:1 compression gives 256 Kbps / sec

A tech spec if you are interested.
http://www.aptx.com/content.asp?site...e=108&pt=apt-X

MP3 at 128kbps 44.1kHz stereo is 90%
compression.

For higher compression (ie. less than 500kbps for stereo), you need to
use a perceptual encoder. These usually use a Fourier Transform or
related algorithm (MP3 uses Modified Discrete Cosine Transform, MDCT)
to convert to the frequency domain, and then a perceptual filter
discards all the frequency components that are not expected to be
perceived. The frequency components, their phases and amplitudes are
then transmitted in the bitstream. (It's a fair bit more complex than
that, of course. I'm generalizing.)

The big drawback with this is the added complexity and delay.

Of course. MLP (as used in DVD-A) was designed to have low decode
latency, so doesn't achieve theoretical maximum compression, which I'm
told (by one of the designers of MLP) that FLAC does. To exceed the
theoretical maximum lossless compression, you need to throw something
away that hopefully won't be noticed.

FWIW theoretical maxiumum lossy compression has a bit rate of zero -
unfortunately it throws away all the audio signal
--
Regards

- replace xyz with ntl

Paul Martin wrote:
In article ,
The Natural Philosopher wrote:
Paul Martin wrote:

For higher compression (ie. less than 500kbps for stereo), you need to
use a perceptual encoder. These usually use a Fourier Transform or
...

That I would accept, BUT the key point in this discussion, is whether
e.g. 'lossless compression' has a fixed data rate, or varies with the
content, and, if so, what the peak to mean ratio is.

It does vary, according to the complexity of the waveform. A delta
encoding would be happiest with a shallow triangular wave.

But, even taking your figures of 500Kbps as a guide, that will fit in
about 50Khz of RF band at a S/N of say 30dB: that's considerably better
than an FM signal occupying say 200Khz of band., which will be almost
unlistenable at that S/N ratio*.

That's purely theoretical. You've not allowed for any error detection
or correction.

I did a little bit ;-)

In theory, you can stuff 64kbps through a phone line (as
that's how it's transported between exchanges as a 64kbps digital
stream) but unless you're plugged directly into the exchange you're
never going to achive it on a real phone line. Radio transmission is
more hostile than cable transmission.

DAB and DVB-T both "waste" bandwidth to achieve robustness in the face
of multipath. (This is the guard interval between symbols.)

Ah but multipath can be catered for with 'echo cancellation' with modern
DSPS.

All you are saying is that current DAB is well below state of the art.

In article , The Natural Philosopher
wrote:
Thanks for the hint..Google revealed what it was..

So you need a channel width of 2 x( 75khz +20khz) for full mono
transmission of the audio band at rated deviation of +- 75Khz..

Or 200KHz. Near enough.

The Wiki note does say that the sidenands are theoretically infinite but
'98% of the energy is within the formula's bandwidth'..

I don't know what you've already found from wiki, but you may find the info
on

http://www.audiomisc.co.uk/HFN/BandwidthBlues/page.html

of interest. Note though, that for simplicity it ignores various real world
factors like phase distortion in filters, and multipath.

Slainte,

Jim

--
Please use the address on the audiomisc page if you wish to email me.
Electronics http://www.st-and.ac.uk/~www_pa/Scot...o/electron.htm
Armstrong Audio http://www.audiomisc.co.uk/Armstrong/armstrong.html
Audio Misc http://www.audiomisc.co.uk/index.html

Jim Lesurf wrote:
In article , The Natural Philosopher
wrote:
Thanks for the hint..Google revealed what it was..

So you need a channel width of 2 x( 75khz +20khz) for full mono
transmission of the audio band at rated deviation of +- 75Khz..

Or 200KHz. Near enough.

The Wiki note does say that the sidenands are theoretically infinite but
'98% of the energy is within the formula's bandwidth'..

I don't know what you've already found from wiki, but you may find the info
on

http://www.audiomisc.co.uk/HFN/BandwidthBlues/page.html

of interest. Note though, that for simplicity it ignores various real world
factors like phase distortion in filters, and multipath.


Well I'll get me coat...that basically summarises much better exactly
what I have been saying.

Slainte,

Jim

In article , The Natural Philosopher
wrote:
Jim Lesurf wrote:


I don't know what you've already found from wiki, but you may find the
info on

http://www.audiomisc.co.uk/HFN/BandwidthBlues/page.html

of interest. Note though, that for simplicity it ignores various real
world factors like phase distortion in filters, and multipath.


Well I'll get me coat...that basically summarises much better exactly
what I have been saying.

OK. Bit warm at present for a coat, though. :-)

With regard to something else I've seen discussed recently...

You can do a simplified estimate of the information bandwidth for standard
stereo FM by assuming it gives 2 channels, each with 15kHz bandwidth (so
needing a minimum of just over 30ksamples/sec per channel - 60ksamples/sec
in total) and assuming a useful dynamic range of, say, 65dB and turning
that into a LPCM bit depth (which would be somewhat less than 16 bits per
sample). Alternatively by using Shannon's Equ directly.

However I am not sure those results are very reliable as there are some
snags.

The obvious ones are that the noise in the demodulated channel isn't white,
and the signals are pre/de-emphasised. These can be dealt with OK.

But distortion also can affect information bandwidth by making patterns
less distinguishable. This is something that simple analysis generally
doesn't cover. And of course FM tends to have quite complex distortion
behaviour.

Books on information theory tend not to dip into this. For much the same
reason as most uni level texts are very careful to only consider simple
sinewave modulation for FM. Beyond that they tend to fall silent. Not even
warning the students about the wilds beyond where monsters may lurk... :-)

In some ways this is analogous with the snags when you try to assess the
real information bandwidth of ye olde vinyl LP. The pre/de-emphasis and
non-uniform noise and peak levels vs frequency are complications that can
be dealt with in obvious ways. But the distortions (particularly those that
produce crosstalk) are harder to assess.

I do wonder if people have fallen into the habit of assuming that FM is
'excellent' partly because when FM stereo was launched actually analysing
some of these problems was virtually impossible, and simple bench tests
with a sinewave into a stereo gen often didn't show them. The result being
many magazine reviews with measured distortions, etc, that were
unrealistically low. IIRC the same was true for sensitivity when fed with a
cable from the generator rather than an antenna radiated with the
background noise from the environment.

From what you have written, I imagine that - like myself in the past - you
have found cases when testing tuners where for any given modulation you can
tweak the tuning or alignment to get very low 'distortion'. The snag being
that you may have to re-tweak when you change the details of the
modulation. Alas, listeners may find this difficult to do when listening to
music at home. :-)

Slainte,

Jim

--
Please use the address on the audiomisc page if you wish to email me.
Electronics http://www.st-and.ac.uk/~www_pa/Scot...o/electron.htm
Armstrong Audio http://www.audiomisc.co.uk/Armstrong/armstrong.html
Audio Misc http://www.audiomisc.co.uk/index.html

In article , The Natural Philosopher
wrote:


Whereas using a small 10KHz signal could often show ghastly distortion
of several percent..

The key is to use modified class AB with bipolars, and run in class A
for small signals, and also to try never to actually switch one side
OFF, as that led to issues with time delay switching it back on:

That was true in early days. However by the early 1980s I found that the
main problem with audio power bipolars tended to be the switch-off time due
to carrier storage in the base region.

That said, during that time the good japanese bipolars had pretty well
cracked this and even with designs that delivered over 100wpc you could get
negligable crossover problems even with bias of the order of 10mA per pair.


A nasty problem of thermal stability was then encountered.

Again, that was certainly a serious problem in early days, particularly if
you used Ge devlces like the beloved sic AL102. :-) But provided you
knew how to work out the stability margin and select the emitter resistor
it wasn't so much of a problem by the early 1980s in my experience. By then
makers could produce devices with consistent specs, unlike in earlier days
when every device seemed like a 'one off special'. 8-]

The adoption of power FETS with much better frequency responses,
negative temperature coefficients and low switch on delays made that
almost a non problem.

Must admit I never liked power FETs for audio as the ones I tried years ago
all liked to hoot at HF, had a habit of current limiting, and shoved
capacitance where I didn't want it. But again this was in the 1980s so I
imagine they rapidly improved and have been fine for some years.

Overall, I'd be happy to use amplifiers that employ either bipolar or fet
if the designer has produced a decent result.

Everyone used feedforward at some point to sharpen up transient
response, and compensate for overall lag. .


Not sure I am "everyone" then. But my memory is fallible. :-)

Slainte,

Jim

--
Please use the address on the audiomisc page if you wish to email me.
Electronics http://www.st-and.ac.uk/~www_pa/Scot...o/electron.htm
Armstrong Audio http://www.audiomisc.co.uk/Armstrong/armstrong.html
Audio Misc http://www.audiomisc.co.uk/index.html

Jim Lesurf wrote:
In article , The Natural Philosopher
wrote:
Jim Lesurf wrote:

I don't know what you've already found from wiki, but you may find the
info on

http://www.audiomisc.co.uk/HFN/BandwidthBlues/page.html

of interest. Note though, that for simplicity it ignores various real
world factors like phase distortion in filters, and multipath.


Well I'll get me coat...that basically summarises much better exactly
what I have been saying.

OK. Bit warm at present for a coat, though. :-)

With regard to something else I've seen discussed recently...

You can do a simplified estimate of the information bandwidth for standard
stereo FM by assuming it gives 2 channels, each with 15kHz bandwidth (so
needing a minimum of just over 30ksamples/sec per channel - 60ksamples/sec
in total) and assuming a useful dynamic range of, say, 65dB and turning
that into a LPCM bit depth (which would be somewhat less than 16 bits per
sample). Alternatively by using Shannon's Equ directly.

However I am not sure those results are very reliable as there are some
snags.

The obvious ones are that the noise in the demodulated channel isn't white,
and the signals are pre/de-emphasised. These can be dealt with OK.

But distortion also can affect information bandwidth by making patterns
less distinguishable. This is something that simple analysis generally
doesn't cover. And of course FM tends to have quite complex distortion
behaviour.

Books on information theory tend not to dip into this. For much the same
reason as most uni level texts are very careful to only consider simple
sinewave modulation for FM. Beyond that they tend to fall silent. Not even
warning the students about the wilds beyond where monsters may lurk... :-)

In some ways this is analogous with the snags when you try to assess the
real information bandwidth of ye olde vinyl LP. The pre/de-emphasis and
non-uniform noise and peak levels vs frequency are complications that can
be dealt with in obvious ways. But the distortions (particularly those that
produce crosstalk) are harder to assess.

I do wonder if people have fallen into the habit of assuming that FM is
'excellent' partly because when FM stereo was launched actually analysing
some of these problems was virtually impossible, and simple bench tests
with a sinewave into a stereo gen often didn't show them. The result being
many magazine reviews with measured distortions, etc, that were
unrealistically low. IIRC the same was true for sensitivity when fed with a
cable from the generator rather than an antenna radiated with the
background noise from the environment.

From what you have written, I imagine that - like myself in the past - you
have found cases when testing tuners where for any given modulation you can
tweak the tuning or alignment to get very low 'distortion'. The snag being
that you may have to re-tweak when you change the details of the
modulation. Alas, listeners may find this difficult to do when listening to
music at home. :-)


What I found was that basically, when I set a tuner up for the then
uncluttered UK spectrum with a 600Khz wide IF strip, it was crap in its
target (German) market because of adjacent channel rejection: Setting it
up for better adjacent channel rejection (400KHz IF) ruined its 'sound'
- at least to my hypercritical ears - for UK. Radio 3 being the source
of choice.

I spent several years staring at the spectra on a spectrum analyser. My
final conclusion was that you were damned if you did and damned if you
didn't. If you left the door open, you got burbles in the stereo from
adjacent channels: If you stripped them out a lot of the stereo went
with it.

Your mentioning of turntables and vinyl also amused me.

3 degrees misalignment from the vertical (just about visible) will
affect stereo separation to the point where its way below what the
cartridge can do. I had MANY problems with acoustic feedback from music
to the cartridge, even on well mounted turntables..how may people
ACTUALLY set up their leads to 'peak' the inherent low pass filter of
the cartridge at the correct frequency..

And don't get me started on loudspeakers..let's by all means have a
ruler flat response on our amps with no tone controls..and minimal phase
shift..and send it to loudspeakers that, even in free space, have a
response like the a ploughed field in the Pyrenees...



Slainte,

Jim

Jim Lesurf wrote:

The adoption of power FETS with much better frequency responses,
negative temperature coefficients and low switch on delays made that
almost a non problem.

Must admit I never liked power FETs for audio as the ones I tried years ago
all liked to hoot at HF, had a habit of current limiting, and shoved
capacitance where I didn't want it. But again this was in the 1980s so I
imagine they rapidly improved and have been fine for some years.


That was merely a question of adapting your driving circuits.

A bit of gate resistance cured the hoot, and you just ended a lot of
gate current to wake them up..

The 80's was when I gave up the whole game as not paying a decent wage,
and turned to computers for income...

Overall, I'd be happy to use amplifiers that employ either bipolar or fet
if the designer has produced a decent result.


Well.. yes.

As with most things 'Hi-Fi' turned from being a high value specialist
product sold to at least the semblance of a discerning public, to a mass
market price sensitive product, where bull**** sold more amps than quality,.

Personally I blame socialism. Too much disposable income in the hands of
people with no taste and even less discretion ;-)

Its the same with cars..until you actually drive a car that has any sort
of handling at all, you cant work out what the fuss is about.

Having said that, I no longer do, nor do I really care much about the
quality of the audio equipment I have.

In the end, I want to listen to the music, not the equipment. Being 'in
the business' ruined the experience of a live rock concert for many years..

In article , The Natural Philosopher
wrote:
Jim Lesurf wrote:

The adoption of power FETS with much better frequency responses,
negative temperature coefficients and low switch on delays made that
almost a non problem.

Must admit I never liked power FETs for audio as the ones I tried
years ago all liked to hoot at HF, had a habit of current limiting,
and shoved capacitance where I didn't want it. But again this was in
the 1980s so I imagine they rapidly improved and have been fine for
some years.


That was merely a question of adapting your driving circuits.

A bit of gate resistance cured the hoot, and you just ended a lot of
gate current to wake them up..

Plus having engineered in another RC rolloff, probably with some included
inductance. :-)

Yes, I did find that stopping the hoot that way was easy. Alas, my
experience at the time was using that time-honoured method (as per grid
stoppers of a pervious age) then fouled up the performance in some other
way. So you then spent your time chasing other problems. There were other
'solutions' but again I decided they just shoved around where a snag popped
up.

I'm sure FETs got better and this became a non-problem. But I decided in
the 1980s that at that time they were more of a pain than bipolars,
particularly when the japanese started producing some really superb audio
power ones. so you could almost forget about secondary breakdown and
carrier storage. And at that time I'd have needed quite a few FETs in
parallel to get me the peak currents I wanted and the bipolars cheerfully
provided. That was the days of apogees, etc. 8-]

The 80's was when I gave up the whole game as not paying a decent wage,
and turned to computers for income...

I went back into academia. If you can't beat em, teach em... er, I mean,
learn more. :-)

Overall, I'd be happy to use amplifiers that employ either bipolar or
fet if the designer has produced a decent result.


Well.. yes.

As with most things 'Hi-Fi' turned from being a high value specialist
product sold to at least the semblance of a discerning public, to a mass
market price sensitive product, where bull**** sold more amps than
quality,.

Personally I blame socialism. Too much disposable income in the hands of
people with no taste and even less discretion ;-)

I tend to point at the dealers who valued an exclusive dealership with a 40
percent markup over actually selling gear that simply did the job with no
hype or an inflated price. But I guess 40 percent of a high price, and no
local competition, was simply too tempting.

And of course 'reviewers' who moved to fantasy island when writing their
articles. :-) The result was a decade or more where anyone who doubted
the magic brands and bull was obviously not to be taken seriously. Hate to
think how much damage that did to many makers and engineers who simply
wanted to produce decent kit, but weren't in the magic circle. Jim Sugden
springs to mind as an example I recall of someone who decided that the bull
made the game one worth walking away from.

The remains are with us still. e.g. Mains cables that cost over a 1000 quid
and have pretty blue lights on them to 'improve the sound', etc. sigh

Having said that, I no longer do, nor do I really care much about the
quality of the audio equipment I have.

It matters a lot to me for the reason you give below...

In the end, I want to listen to the music, not the equipment.

That's why I still care about the audio gear I use, and that I should use
it in an optimum way. It allows me to enjoy the results more. But I do that
in ways that do make engineering sense to me. Not by buying eyecandy or
jewellery for audiophiles. :-)


Being 'in the business' ruined the experience of a live rock concert for
many years..

Fortunately I realised after a few years that I was focussing on things
like watching waveforms on a scope or trying to hear the quack from LS
cones to find problems, not listening to music. Once I'd realised this I
changed tack. I now rarely buy equipment and mostly just enjoy the music.

Most of the main gear I use for audio is decades old. Still works fine. And
unlike a lot of modern kit is easy to fiddle with if needed.

I do still experiment and try to learn more, though. Most recent example
being a look at using Linux boxes for playing audio. I was not surprised to
find some problems, but pleased that they could be sorted out OK. If anyone
is interested, the results are here

http://www.audiomisc.co.uk/Linux/Sou...Computing.html

However I do wonder how many people are listening to systems that are
fudging up the sounds without them knowing this or that they can be
improved. I was able to generate and measure test files to find the
problems. But I guess most people can't/don't do this, and then presumably
either think it is OK or if not, may blame something else.

Slainte,

Jim

--
Please use the address on the audiomisc page if you wish to email me.
Electronics http://www.st-and.ac.uk/~www_pa/Scot...o/electron.htm
Armstrong Audio http://www.audiomisc.co.uk/Armstrong/armstrong.html
Audio Misc http://www.audiomisc.co.uk/index.html

In message , Jim Lesurf
writing at 15:28:44 in his/her local time
opines:-

As with most things 'Hi-Fi' turned from being a high value specialist
product sold to at least the semblance of a discerning public, to a mass
market price sensitive product, where bull**** sold more amps than
quality,.

Personally I blame socialism. Too much disposable income in the hands of
people with no taste and even less discretion ;-)

I tend to point at the dealers who valued an exclusive dealership with a 40
percent markup over actually selling gear that simply did the job with no
hype or an inflated price. But I guess 40 percent of a high price, and no
local competition, was simply too tempting.

And of course 'reviewers' who moved to fantasy island when writing their
articles. :-) The result was a decade or more where anyone who doubted
the magic brands and bull was obviously not to be taken seriously. Hate to
think how much damage that did to many makers and engineers who simply
wanted to produce decent kit, but weren't in the magic circle. Jim Sugden
springs to mind as an example I recall of someone who decided that the bull
made the game one worth walking away from.

The remains are with us still. e.g. Mains cables that cost over a 1000 quid
and have pretty blue lights on them to 'improve the sound', etc. sigh

Having said that, I no longer do, nor do I really care much about the
quality of the audio equipment I have.

It matters a lot to me for the reason you give below...

In the end, I want to listen to the music, not the equipment.

That's why I still care about the audio gear I use, and that I should use
it in an optimum way. It allows me to enjoy the results more. But I do that
in ways that do make engineering sense to me. Not by buying eyecandy or
jewellery for audiophiles. :-)


Being 'in the business' ruined the experience of a live rock concert for
many years..

Fortunately I realised after a few years that I was focussing on things
like watching waveforms on a scope or trying to hear the quack from LS
cones to find problems, not listening to music. Once I'd realised this I
changed tack. I now rarely buy equipment and mostly just enjoy the music.

Most of the main gear I use for audio is decades old. Still works fine. And
unlike a lot of modern kit is easy to fiddle with if needed.

I do still experiment and try to learn more, though. Most recent example
being a look at using Linux boxes for playing audio. I was not surprised to
find some problems, but pleased that they could be sorted out OK. If anyone
is interested, the results are here

http://www.audiomisc.co.uk/Linux/Sou...Computing.html

However I do wonder how many people are listening to systems that are
fudging up the sounds without them knowing this or that they can be
improved. I was able to generate and measure test files to find the
problems. But I guess most people can't/don't do this, and then presumably
either think it is OK or if not, may blame something else.

Slainte,

Jim

Ah the Armstrong 223! I had one of these, and I well remember buying the
stereo decoder add-on and installing it - thus becoming one of the
pioneers in hearing the Northampton stereo 'birdies'.

Funny how 10 (genuine, RMS) watts could practically make your ears bleed
back then, and my home cinema, today, alleges it puts out 850.

But the Rogers Cadet Mk III could certainly drive that pair of 8in
Wharfedale RS/DDs, in their kit cabinet, and the Garrard SP25 with its
cheap Goldring cartridge did the business.

That was entry level hi-fi back then, IIRC; but the gap between that and
the 'finest' radiogram was a yawning chasm, populated by more coloration
than a Disney cartoon.

And the finest radiogram was dearer than that system - though it did
have space to store some LPs at least.

For me, it was never 'money no object' - it was always 'how cheap can I
get good sound?'

What did for the old-style 'hi-fi', IMHO, was a closing of the gap - to
the point where the second-best system in our house, a £180 JVC micro
setup, needs close A/B listening to distinguish it from the Arcam CD/
Audiolab amp/ Spendor BC1s setup I now have; at 'normal' listening
levels at least.

So no wonder hi-fi went three ways - cheap mug's-eyeful stuff that is no
better than it should be; £kkk bling that says 'look at me' instead of
'listen to me'; and the honest but narrowing middle ground where the
good stuff still wins out over the mass-produced - but you have to
concentrate to hear the difference.

Hell, even the best MP3 players sounded like they were underwater until
a few years back - but my iPod Touch (with Sennheiser PX200s, of course)
beats my last-generation Sony Walkman cassette player into a cocked hat
(and doesn't skip like my portable CD player, even though that may be a
little better, objectively and subjectively, if I can keep it still
while I play it).

But what about the modern, subjective, 'hi-fi' review? I have to confess
they send me screaming as being just too unscientific and sometimes
outright bullsh; but even back in the old days, I knew the 'B&K graph'
reviews in HI-FI News were missing something, when the graphs from the
Shure 75EJ stylus looked just like those from the 75ED - yet two seconds
was all it took to tell them apart when you swapped them over.

And it was pretty 'subjective' when I took the Cadet to the local hi-fi
repairer and said that it sounded sort of 'like a lorry struggling up a
hill instead of the car cresting it that it always used to' - and he
found the main power supply capacitor had failed, which sounded like a
pretty plausible explanation.

Do cables make a difference? Sure they do. With Litz cables on my Luxman
M300, whacking the treble over to full would make it oscillate rather
badly.

Do they ever make a *good* difference? Pass. But to put the £30 Tesco
DVD player with HDMI, on our 20in Bravia TV in the bedroom, I wasn't
going to spend the same again on an HDMI cable (let alone the twice
that, like the QEDs that feed BluRay to the 46in behemoth in the
lounge), so I bought a £3.50 one.

Just for grins, though, I tried it in the lounge. None of us could see
any difference, even on my THX demo disc (though in fairness that isn't
BluRay, but other discs we tried were).

OTOH, £20 SCARTs showed a visible difference over in-box freebies.

But none of this was double-blind of course, so don't take my word for
it :-)


--
Roy Brown 'Have nothing in your houses that you do not know to be
Kelmscott Ltd useful, or believe to be beautiful' William Morris