Multigig modem/router mod project

Xymox

Active Member
Apr 17, 2019
95
52
25
58
I am going to do a thread with detailed info/pics/measurements as I mod a Arris S33 Multigig cable modem and a Mikrotik Multigig router .

The residential cable world is now going multigig turning on 10G. Speeds above 1Gbps will soon be common. 10G is the future for homes. It comes with vastly lower latency and jitter too. Most super high end audio users will want to have the fastest tier of service which will not be possible without a full change out of modem, router and WAPs. While the standard residential user will buy a all-in-one modem/router/WAP device this setup wont work well for high end audio as custom tweaked routers and modems are much better sounding.. SO.. My project here will be a exploration into the new multigig world.

Current devices in use in high end audio have 1 Gbps ports. None of this will work in the multigig world. Also router performance needs to be carefully examined for specs for thruput with NAT and Firewall enabled in various packet sizes. The Ubiquiti product line is not suitable for Multigig use, with the exception of thier $1800 top end device. However this has not been modded or used in high end audio. It is also not state of the art and performance limited VS Mikrotik products. So Ubiquiti is no longer viable.

Audio streaming tho is low speed. So your switches, SFP, optical, networking cables on your clean networking setup for audio are fine and will be for a long time yet. The switch will connect to the multigig router via a 100/1000 Gbps ethernet port or optical via SFP. So my focus is on the router and modem as those must be replaced if you want multigig in the house for normal use on all your devices.

The devices I have chosen were well considered as a base for mods. The Arris S33 modem is the multigig evolution of the Broadcom BCM3390 modem chip. Same chip as the SB8200 but a better multirate ethernet driver chip was hooked to the BCM3390 which provides a 2.5Gbps ethernet. The router has 10 SFP+ ports which can take multigig SFP+ Ethernet modules converting a SFP+ into a 10/100/1000/2500/5000/10000 Ethernet port. It is FULLY modular and allows you to configure the router to have any SFP/SFP+/SFP28 module you want in any combo and configuration. The router is clean and well laid out for mods. Both the modem and router can be easily adapted for linear supplies.

This will be a evolving project in the months ahead as I play and listen and play. This first pass on the devices will be just experimentation and basics. Eventually I will settle into a good set of mods. These first passes will look a bit ruff and not finished pretty, its experimentation. Later I will clean it all up and do a final version.

I don't think I want to make these products. I might. We shall see. Hopefully I can spur some of you to do your own work on these devices and go beyond anything I do. So my intent is to supply as much info as I can...

I got my new modem yesterday and the router does not get here for a few days yet.. So I will start with the modem...

10G is the future and all of you will want it. You can only have one cable modem and router, so, those need to be multigig..

 
Last edited:

Xymox

Active Member
Apr 17, 2019
95
52
25
58
This is the modem before mods.. This zip file has a lot of pics..

I have another side gig. I test modems for various ISPs and vendors. I created the bad modems web site. I also am class representative in a class action suit involving the Intel Puma modem. I developed tests now std in the industry for testing modem performance. I have tested most every modem for performance specs like jitter. I have a fair bit of experience with cable modems. I also do these same tests on routers.

The Arris S33 is the best modem I have tested. It is based on the Broadcom BCM3390 chip and its performance is well established. The S33 tho ads a far more powerful multirate Ethernet chip. So my tests for jitter with some quick prelim tests show its about 10% lower jitter then a SB8200 which is the best until now. The build quality is excellent. Component quality is excellent for a modem. Thermal management is reasonable. It is very similar in circuit topology to the SB8200. The part layout makes it pretty easy to mod. The case is a new design for Arris and takes away some room inside the case for big caps tho. The board could be removed and placed in a much roomier case tho.

The modem is very typical of all modems. A SOC, BCM3390, that has nearly everything on it except for RF and power supplies. It feeds a RF transmitter and listens to a receiver. These are NOT normal RF sections and are very wideband and have all sorts of very state of the art RF voodoo. OFDM and QAM modulators/rcvrs. The multigig ethernet port has its own controller I will get the part number later. The 1Gbps port is driven directly by the BCM3390. These have std isolation transformers.

If you want to go on a deep dive into what makes DOCSIS 3.1 cable modems tick.. This is a good paper

The incoming 12V is fed to various supplies and the incoming 12V has no filtering or any cap to speak of.

The case is unremarkable typical flame retardant plastic. It is easy to get apart tho with 2 screws under the feet on the bottom.

It has 8 "buck boost" style voltage regulators. I hate these. They are in every digital device these days and are really just switching power supplies. They use PWM to regualte voltage outputs. These are highly noisy and have a list of limitations. The BCM3390, linear sections, RF sections and Ethernet drivers are all fed from these switching regulators. These will be my first focus.

These internal PWM voltage regulators produce 1.5Mhz as the primary and zillions of harmonics that vary with regulation and load. You cant use electrolytics at these frequencies. The best approach are huge super low ESR Vishey/Sprague tantalum caps. The main cap for the 12V can be a nice audio cap tho. The idea is to vastly clean up the high freq noise on the chip rails as the RF and Analog sections of a cable modem REALLY respond nicely to clean supplies. These switching regulators universally call out a 10uF output filter cap and at least that on the input side. Because cable modems MUST be made cheaply, and ultra low ESR parts are required and expensive,, they never put good amounts of filtering on these. They do just enough so it works. They literally cheapen it up even after design leaving off some uF in the designs.

if your interested in how these PWM regulators work this is a typical one.

I don't fool around with these supplies. I never have. I have modded modems for 10 years.. All these power rail regulators will get a 100uF/220uF ultra low ESR tant cap on the output side and at least 100uF on the input side.. Right now on the S33 it varies but they are all less then 0.1uF. So this will be a increase of 100 to 220 times in uF. As I have done this many times I know it will result in a dramatic drop in noise across the whole bandwidth dc-3Ghz. I have a real spectrum analyzer and I look at this.

The clocks are not a off the shelf frequency for a OXCO being 50Mhz and 54Mhz. I will have to ponder this.. I will measure the phase noise of them. Cleaning up the supplies will help a little for these.

The ethernet jacks are shielded and the shield is grounded to the - power supply rail and to the coax. So you do NOT want to use a shielded ethernet to connect this to a router as it would propagate the coax ground to the router and maybe to your gear.

So... Lets jump in... First up order parts.. I have my order placed with Mouser, but that will take some days to get here yet..

I have circled the modem power rail regulators in white. The red line shows the caps on the output side, yellow is input. They use a number of caps with different sizes to produce a good low ESR and regulation response at the cheapest cost. These are good caps tho for this use and well tuned to the circuit. So I will leave those and just add the tant cap to them.



This shows the side of the board and the other regulators and the 12V input.... NOTICE they did not even put a cap that is in the design for filtering on the input supply, purple. I will drop 1360uF of tantalum there.

On all these pics on the input ( yellow ) side they left off caps that were in the design. Again cost savings.



So that is the basics of my Mouser order and first exam..
 
Last edited:

Xymox

Active Member
Apr 17, 2019
95
52
25
58
I am going to add a small quiet fan to mine. Its not "required" but I feel all modems leave out a fan and don't mind cooking the electronics. I will put my modem away from the audio gear.. So a tiny bit of fan noise wont bother me..

It looks like it will fit very nicely into the top. I will cut the top plate to fit around this fan.. It will be quite nice but will protude some above the case.

 
Last edited:
  • Like
Reactions: audioblazer

Xymox

Active Member
Apr 17, 2019
95
52
25
58
I forgot to mention voltages.. Not that its that important, but, i needed them for ordering caps..

8 supplies..

+12 1.5A from outside. This is used as input for all the requlators.

+5
-5
+1.8
+3.3
+1.08
+1.5
+.95
+.95

 
Last edited:
  • Like
Reactions: audioblazer

Xymox

Active Member
Apr 17, 2019
95
52
25
58
The router will have the same kind of switching regulators and I will do about the same things. The router however will also get careful attention to some caps for a audio focus..

If your using a linear supply with a modem or router that will make a difference, however, most of that is being lost because of the switching regulators feeding all the internal parts..

REMEMBER.. These are just first stabs and kinda experimentation. This project will evolve.
 
  • Like
Reactions: audioblazer

Xymox

Active Member
Apr 17, 2019
95
52
25
58
I have done some more detailed testing.. I need some specs before mods so I can compare when done. I also want to get some good numbers to start with..

This is a standard test I run. I use Ping Plotter Pro. I use non-standard settings. This test is ICMP Ping. I target my ISP CTMS. This is the device on the other side of my cable modem at the ISP. This is a real world test but is obviously influenced by my ISP.

This is 100 pings PER SECOND. Its a very high temporal rate and reveals even tiny variations in jitter and latency. This is CRITICAL for audio streaming. This high temporal rate reveals even tiny variations in latency and jitter.

I use these same tools and high temporal rate testing with routers. I test THRU routers all locally. This kind of testing can show jitter performance of a router or switch. I also pass high rates of data thru a device at the same time to see how latency and jitter are effected.. I will get to all that once I get to the router..

This shows a 6.8ms avg latency and a 1.2ms jitter avg over 10 mins with 12,000 tests in 10 mins. This is really good.. Of course this is effected by my ISP link and the load on my CMTS and devices along the way.. I test lots of devices this way on my same link and this is the best i have seen. I think I am now seeing ONLY ISP load. I don't think the modem is contributing much at all. So this is really great performance.

 
  • Like
Reactions: audioblazer

cat6man

Well-Known Member
Feb 7, 2013
344
198
385
west of NYC, east of SF
I have done some more detailed testing.. I need some specs before mods so I can compare when done. I also want to get some good numbers to start with..

This is a standard test I run. I use Ping Plotter Pro. I use non-standard settings. This test is ICMP Ping. I target my ISP CTMS. This is the device on the other side of my cable modem at the ISP. This is a real world test but is obviously influenced by my ISP.

This is 100 pings PER SECOND. Its a very high temporal rate and reveals even tiny variations in jitter and latency. This is CRITICAL for audio streaming. This high temporal rate reveals even tiny variations in latency and jitter.

I use these same tools and high temporal rate testing with routers. I test THRU routers all locally. This kind of testing can show jitter performance of a router or switch. I also pass high rates of data thru a device at the same time to see how latency and jitter are effected.. I will get to all that once I get to the router..

This shows a 6.8ms avg latency and a 1.2ms jitter avg over 10 mins with 12,000 tests in 10 mins. This is really good.. Of course this is effected by my ISP link and the load on my CMTS and devices along the way.. I test lots of devices this way on my same link and this is the best i have seen. I think I am now seeing ONLY ISP load. I don't think the modem is contributing much at all. So this is really great performance.


very interesting!

could you share results for other, not as optimized, configurations?
how bad was it with off the shelf hardware and settings?

and a followup question: have you been able to correlate any SQ benefits and to quantify what the jitter was before and after you heard the SQ improve?

i can do a similar test on Linux with the ping command (don't know about windows/mac)

sudo ping -c 60000 -i .001 192.168.1.156 > pingtest.log

# ping at 1ms (.001) interval, count of 60000 pings =1minute, save to log file, sudo required for less than 0.2 interval

192.168.1.156 is my internal address for my LMS music server

Pinging my internal machine had an average latency of 0.254ms and a rms jitter of 0.055ms.
Pinging tidal.com just now had an average latency of 4.47ms and a rms jitter of 0.865ms, which seem to be better than your ISP pings.

Any thoughts?

it would be interesting to find the ip address of my fios connection at the ISP and the IP address my Tidal account is working with (as opposed to tidal.com)

on the other hand, couldn't flooding pings get your IP address banned?
 
Last edited:

Xymox

Active Member
Apr 17, 2019
95
52
25
58
The normal ping command can't do 100 pings per second and graph results. Ping Plotter is the best tool for this. I have used every tool and on the *nix side prefer FreeBSD rather then Linux. Also you need good hardware to run PingPlotter on for this kind of testing. I have a i7 X990 overclocked to 4.2Ghz. I prefer this CPU over newer ones. Fairly crazy MB. Overclocked serious ram. 4 RAID 0 Samsung SSD. I use a Intel server ethernet card and also a Intel SFP+ server NIC. I built this machine specifically for doing these ping plotter high temporal rate plots.

Oh yea, as I go I will post lots of charts.

I have never really focused on the SQ of the router and modem before. Those are for streaming over the internet and that sound quality was not that important to me. The quality of the clean network segment that I always setup is done with a few SoTMs, regens, galvanic isolation with linears and SFP optical and is my main focus for sound quality between the server and renderer. For that I use my ears as the test equip does not give a complete picture of the freaky voodoo that seems to be going on with SQ on a packet based music system. I use pingplotter tests just to be sure nothing untoward is going on. Fine tuning has to be done by ear. I have a number of high end systems I can check gear with. the one I will use to evaul this project will mostly be done on a system with a Taiko Extreme, CH Precision C1d + X1's + T1 with external GPS. Various SFP modules and various ethernet cables.

BUT SQ evaul is a bit off yet. I know what to do to clean things up to start with. The nasty voltage regulators need cleaning up on the modem and router as step one. The nasty pervasive RF noise from them that gets into everything needs to be eliminated. The power rails to the chips needs to be stabilized. Once i get all that done I will measure things again and use the spectrum analyzer looking at RF and really decide how to proceed towards SQ improvements.

Honestly I am skeptical. The internet based music sources have so much jitter and junk before they reach the modem its hard to believe a modem and router will matter. However LOTS of people here say it does matter. A lot in fact.. So, why not...

I also want to take a deep dive into the physical layer on Ethernet. i want to look at it from a noise/spectrum point of view. A ethernet cable is, at its core, a analog signal that conveys digital data. I want to look carefully at this analog signal.
 

Xymox

Active Member
Apr 17, 2019
95
52
25
58
My badmodems site has plots and tests of some other devices. The "What's the issue" section has some plots. Ive done a lot of testing and plots over at DSLReports. Mostly on the Intel Puma thread which is where all my testing began 4 years ago.
 

Xymox

Active Member
Apr 17, 2019
95
52
25
58
I should also point out that Ping Plotter can do the same testing using TCP or UDP not just ICMP Ping. Real testing should be done using the same protocol your intended data will be in.

Also. I use a PC. I don't think you can run these tools properly inside a VM on say a Apple. The VM would introduce jitter into the plots and most likley bork at low intervals ( high temporal rates )..

To get Ping Plotter doing 100 pings per second you have tp type in .01 into the interval box and press enter.
 

Xymox

Active Member
Apr 17, 2019
95
52
25
58
My router was delayed because of Christmas apparently. Fedex has rescheduled for monday.. My Mouser order for caps should process and ship on monday so I should get tant caps for the modem late this week...

Hopefully I will have the router tomorrow and will do a video to show why i picked this router in terms of its modular aspects and inside layouts. I should be able to get a Mouser order together for it tuesday.

I am impatient. I want to start into this and start having fun..
 

Xymox

Active Member
Apr 17, 2019
95
52
25
58
As expected the power supplies are a huge mess with RF... This is a look up to 1Ghz.. Big mess.. Its true on all the supplies and there is all sorts of stray RF all over everything including the Ethernet outputs.. No surprize.. But some good base measurements..

AND this is at idle.. No coax connected, no ethernet.. Modem is not connected... No doubt this will get worse when in operation..

SCREN065.GIF
 

Xymox

Active Member
Apr 17, 2019
95
52
25
58
Can you say switching reg noise ? This is a 2Mhz sweep.. On the 3.3V power rail to the BCM3390. What a mess.. 30db of RF spikes.. These are what I am going to address.. This mess ends up in everything.. This **** also radiates to other PCB traces and just soaks into the whole unit. The giant tantalum caps will knock all this down to near nothing..

SCREN068.GIF
 
Last edited:

Xymox

Active Member
Apr 17, 2019
95
52
25
58
I zoomed way in and looked at one of the spikes.. This is 1.343Mhz center, 32khz screen width. 100hz rez BW.

You can see this spike has harmonics.. The smaller spike on the R/L sides. So this is just a huge RF noise **** storm..

None of this is unexpected.. VERY typical. I see this on these rails on everything that uses this type of voltage regulation which includes all the switches and routers used in high end audio. I don't see many mfgrs addressing this mess. They put fancy caps before all these regulators, but,, the mess remains..

35 DB spike.. This is one of a zillion of them.. Each of them with harmonics. They are all related to the PWM being used in the regulators..

SCREN069.GIF
 

cat6man

Well-Known Member
Feb 7, 2013
344
198
385
west of NYC, east of SF
The normal ping command can't do 100 pings per second and graph results. Ping Plotter is the best tool for this.

i'm curious...........what do you see and use from the plots that gives you useful information beyond min/mean/max/standard deviation?
 

Xymox

Active Member
Apr 17, 2019
95
52
25
58
i'm curious...........what do you see and use from the plots that gives you useful information beyond min/mean/max/standard deviation?
Mainly the look of the plot. How smooth is it. Any odd spikes. Devcies that process packets using Software Defined Networking always have a magnitude higher amount of jitter. It also will have odd patterns in the plot. As the CPU is used to process packets in SDN based devices you can see the CPU do things not related to the packet processing in the plots. Look up table depletion and other things show up in plots for SDN devices.. A good switch/router/modem will seem uneffected by almost anything you do to it. For example, you test thru it from port 1 to port 2. Nice flat plot. Then you run 10Gbps thru port 3 to port 4 WHILE your still testing thru port 1-2. This will show how much jitter is induced in the port 1-2 connection from traffic on 3-4... With SDN devices, and devices that don't offload packet processing, this same test will be brutal.. The 1-2 connection will be radically effected by things going on anywhere else.. You dont want that.. Jitter is introduced this way...

So.. I look for a flat line that is uninfected by other things..

I don't really even look at the numbers, just kinda as a reference.. Its the high rez plot that counts..
 

cat6man

Well-Known Member
Feb 7, 2013
344
198
385
west of NYC, east of SF
Mainly the look of the plot. How smooth is it. Any odd spikes. Devcies that process packets using Software Defined Networking always have a magnitude higher amount of jitter. It also will have odd patterns in the plot. As the CPU is used to process packets in SDN based devices you can see the CPU do things not related to the packet processing in the plots. Look up table depletion and other things show up in plots for SDN devices.. A good switch/router/modem will seem uneffected by almost anything you do to it. For example, you test thru it from port 1 to port 2. Nice flat plot. Then you run 10Gbps thru port 3 to port 4 WHILE your still testing thru port 1-2. This will show how much jitter is induced in the port 1-2 connection from traffic on 3-4... With SDN devices, and devices that don't offload packet processing, this same test will be brutal.. The 1-2 connection will be radically effected by things going on anywhere else.. You dont want that.. Jitter is introduced this way...

So.. I look for a flat line that is uninfected by other things..

I don't really even look at the numbers, just kinda as a reference.. Its the high rez plot that counts..

got it............with the plot you see things in real time, so you can make adjustments on the fly. the numbers probably contain much of the same information but not the real time feedback of a plot.
cool.
 

Xymox

Active Member
Apr 17, 2019
95
52
25
58
got it............with the plot you see things in real time, so you can make adjustments on the fly. the numbers probably contain much of the same information but not the real time feedback of a plot.
cool.
Yea after you have tested as many devices as I have and done a stupid number of AB plots you can just look at a plot and know a lot about what is going on..
 

About us

  • What’s Best Forum is THE forum for high end audio, product reviews, advice and sharing experiences on the best of everything else. This is THE place where audiophiles and audio companies discuss vintage, contemporary and new audio products, music servers, music streamers, computer audio, digital-to-analog converters, turntables, phono stages, cartridges, reel-to-reel tape machines, speakers, headphones and tube and solid-state amplification. Founded in 2010 What’s Best Forum invites intelligent and courteous people of all interests and backgrounds to describe and discuss the best of everything. From beginners to life-long hobbyists to industry professionals, we enjoy learning about new things and meeting new people, and participating in spirited debates.

Quick Navigation

User Menu

Steve Williams
Site Founder | Site Owner | Administrator
Ron Resnick
Site Co-Owner | Administrator
Julian (The Fixer)
Website Build | Marketing Managersing