Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Weaver]

[PhilipD]

Hi

First and foremost Hi-Fi style expensive components are snake oil a lot of the time, especially true in this digital age.

You couldn't make a souped up modem/router on the component level any better than is already out there as a standard mass produced device because of the frequencies involved.  Keeping signal paths very short and well screened or correctly positioned earth planes can only be done with surface mount components and multi-layered circuit boards, and designs can only be validated using expensive testing gear.

The ZyXEL does include an additional filter stage very close to the input, and you could perhaps do the same with another device by breaking a couple of solder tracks and adding one in.  However this filter may need to match with other components to work correctly so could have detrimental effect if added to another device not designed with it in mind.

Some potential improvements could be made by replacing capacitors to newer ones if the device is already a little old as these age quickly and then start to raise the noise floor so SNR drops due to internal noise, but that would only bring it up to as new performance levels, and replacing them isn't always easy due to the multi layered boards.

You could go to a Chinese manufacturer and have some input into it and design your own router/modem from scratch using top quality components etc, but unless you want and know you can sell tens of thousands of them and can prove you can pay for the costs up front, just making a few hundred will be extremely costly.

So I would say yes it is possible, but not cost effective and the risk is the performance may be no better anyway.

Regards

Phil

[Weaver]

> You could go to a Chinese manufacturer and have some input into it and design your own router/modem from scratch using top quality components etc

I wonder what you could gain by using better components. More reliability, reproducibility isn't the DSL equivalent of Hi-Fi (high performance).

A couple of examples:

The first Quad CD player, iirc, was made out of Philips components that were cherry-picked, the one DAC chip in a hundred that measured the best for linearity was chosen, and these cost a lot for Quad to buy. So that was one interesting approach, Darwinian design.

My own old Meridian DAC used two DACs per channel, with the input to one device digitally negated and then the difference between the two analog outputs was taken using a differential amplifier. So throwing more hardware at the problem and more cash.

A couple of success stories concerning how to take something ordinary and turn it into something special rather than having an all-new design which would be horribly expensive and not necessarily successful given that the existing component manufacturers new their business and had done a really good job.

However it isn't immediately obvious to me how this kind of thinking could translate to the problem of improving a DSL modem.

[Weaver]

Better software is one thing to look at, which is not an option because that presumably means getting PhDs with even longer beards and the ones you had were already the best to be had.

I wonder if more can be done with faster processors. More cores might be useful. But it isn't clear to me how faster CPUs could help quality. The CPU is either fast enough for the program chosen or it isn't, the only option that occurs to me is using a different algorithm or using a more lavish variation on an existing one, for example obtaining more accurate values by doing a few more rounds of a computation. Sampling an analog input at higher res, more data points, is possible with a faster clock speed.

But looking at the analog bits of a modem seems very interesting to me as this is where costs might be cut to the bone and then further, compromising the results. I wonder about something simpler like going for a metal case instead of a plastic one?

Way beyond my expertise.

[aesmith]

Whether it's components, design or software, I can say for sure that the many Cisco 831, 877, 887 routers that we've deployed are far more reliable than typical consumer grade DSL routers.   Not so much in terms of hardware failure, but in terms of not getting their knickers in a twist and needing to be rebooted, or not needing to be rebooted to clear transient line faults.

[PhilipD]

Hi

I wonder if more can be done with faster processors. More cores might be useful. But it isn't clear to me how faster CPUs could help quality. The CPU is either fast enough for the program chosen or it isn't, the only option that occurs to me is using a different algorithm or using a more lavish variation on an existing one, for example obtaining more accurate values by doing a few more rounds of a computation. Sampling an analog input at higher res, more data points, is possible with a faster clock speed.

It isn't the CPU running the ADSL/VDSL line modem part though.  The ADSL/VDSL part is dedicated silicon, we can't improve on what is provided without becoming a manufacturer of LSIs, all we have access to is various tweaks, usually a sanitized collection of them via software and using Telnet.  There will be more low level tweaks available via software or changing registers but they will not be easily accessible, will need an in-depth knowledge of the technology, and likely any changes would break compatibility with the modem the other end.

Also if improvements could be made, then the manufacturers would have done just that already and are doing so, VDSL is just an improvement over ADSL, G.Fast is another improvement again, G.INP and Vectoring are further examples of improving the technology. 

But looking at the analog bits of a modem seems very interesting to me as this is where costs might be cut to the bone and then further, compromising the results. I wonder about something simpler like going for a metal case instead of a plastic one?

A metal case may or may not improve things, it's all about designing all that in at the beginning.  Usually a metal case is used for robustness, perceived value of the product, heat dissipation perhaps, and often to stop it interfering with other devices.  However any components susceptible to interference will already be screened off very tightly by can's on the circuit board.  Don't forget the ADSL/VDSL signal travels in un-shielded twisted pair cable and picks up tonnes of noise, so putting a metal case over the modem is shutting the stable door after the horse has long gone.   

Regards

Phil

[Weaver]

@PhilipD  how much is done in completely hard-wired silicon then, not processed by software? Forgive my total and utter ignorance.

[JGO]

Treating a modem as a radio receiver, it wouldn't be difficult to go flat out for sensitivity and pull a weak signal out of noise resulting in a speed increase BUT the increased sensitivity makes the modem more prone to overload by interference ! You can't have jam on it ! This problem has been appreciated since the earliest days of radar in the '30s

Discussions of some modems a year or two ago suggested there is an appreciable difference in sensitivity.

[PhilipD]

@PhilipD  how much is done in completely hard-wired silicon then, not processed by software? Forgive my total and utter ignorance.

Hi

All of the ADSL or VDSL part of it is done by hardware. It used to be separate chips a few years ago but now typically its all in the same package but still is its own subsystem.  It might run software and be upgradeable but its very specialised microcode and would be closed source.

The CPU runs higher level functions such as the Web UI, logins, configuration, DHCP, file sharing etc. Typically routing and firewall functions are off loaded also to dedicated LSIs.  This is why usually the CPUs are pretty low speed, as nothing much more is required. 

Regards

Phil

[Weaver]

@PhilipD  it must be one hell of a challenge implementing that lot in hardware.

[Alex Atkin UK]

The VDSL modem is dedicated silicon but exactly how it functions is controlled by its firmware which it loaded up when the modem boots.

This is why Openreach were able to add G.INP when the modems originally did not support it.

As already pointed out, most performance problems are caused between the modem at the other end of the connection and your home, because its unshielded cable in various states of disrepair.  As such a more sensitive modem can perform worse on one line, better on another.  There isn't really a way you can know what would be best for your given line because of all the variables out of your control.

[skwark]

If you have the tech knowledge and business acumen then you could always crowd-fund it and get your pre-orders before too much outlay. There's no pure modem product that I'm aware of but Turris Omnia Router is in a similar vein and that's been quite successful so far, but then has some serious engineering backing https://www.indiegogo.com/projects/turris-omnia-hi-performance-open-source-router#/