POWER LEVELS & SNR: A TECHNICAL PRIMER (Nov 2011)

[Sephiroth]

POWER LEVELS & SNR: A TECHNICAL PRIMER (Updated November 2011)

PART A - APPLICABLE DOCSIS STANDARDS

INTRODUCTION

There are no precise numbers for power levels although you can be sure that the numbers offered as optimal here will not be the cause of any problems.  This article partially explains how power levels are applied across the different network components.  It also suggests that if your power level and connexion are stable, then power levels won't be your problem.  Conversely, the article explains what the effects of out-of-range power could be if speeds are poor or the connexion is unstable.  The article is EuroDOCSIS 3 oriented acknowledging the 30/50/100 meg tiers.

There are three parts to this article:

PART A:  Deals with the EuroDOCSIS standards applicable to VM's Upstream & Downstream service components.

PART B:  Deals with Upstream transmission and what happens under various circumstances.

PART C:  Deals with the Downstream and what happens under various circumstances.

Acknowledgement is given for the contributions made on the VM Forum by Horseman, Ignition, Lee_G (Team Member), Cabsandy, Apcyberax, Canveyboy and Paultechy.

CABLE MODEM NOMINAL SPECS

DOCSIS 3 POWER LEVEL STANDARDS (CM-SP-PHYv3.0-I08-090121)

VM's DOCSIS network is mix of DOCSIS (6 MHz downstream channel bandwidth) and EuroDOCSIS (8 MHz bandwidth) areas inherited over the NTL/Telewest merger. The main tables here will focus on EuroDOCSIS; for those who need to know about the marginal differences in power tolerances also marginally different - EuroDOCSIS power limits are as per the table below; while DOCSIS specs are nominally -15dBmV - +15dBmV on both 64QAM & 256QAM..

Without making this too complicated, the specs for EuroDOCSIS set a power range for a single channel & for a bonded group.

dBmv @ 6952 Ksyms/sec	Per DS Channel	Bonded total power
DOWNSTREAM 64QAM	-17 to +13 dBmv	< 33 dBmv
DOWNSTREAM 256QAM	-13 to +17 dBmv	< 33 dBmv

On the matter of aggregated downstream power. It's not a case of just adding up the dBmv listed on your 4 or 8 channels.  The (logarithmic) dBmv scale is relative to 0dBmv and a +3 dBmv increase in power level doubles the power arriving at the SH (you can Google all that).  Let's say that the power level on a single channel is 0 dBmv.  Now let's double that to 2 channels so you aggregate power in dBmv terms is +3 dBmv.  Make that 4 channels and it's +6 dBmv.  Your SH currently sees a spread of 16 DS channels (each 8 MHz wide) in your set from which you can actually use 4 or 8 according to the valid BPI+ key associated with your lease.  But all that stuff hits your SH.  16= 2^4 so the aggregate downstream power is at least 4 x 3 dBmv = 12 dBmv.  You will bust the EuroDOCSIS 3 limit of 33 dBmv if the average power across all downstream channels exceeds 22 dBmv.  A clue to this would be that what you see averages c. 23 dBmv. (It's a bit more complicated than that when more channels are introduced; also higher frequencies bring down the average power - but that's not happening now to an extent that affects this article). 

Single channel Upstream	TDMA – DOCSIS 1 (@ 2560 Ksyms/sec)	ATDMA – DOCSIS 2 (@ 5120 Ksyms/sec)
UPSTREAM QPSK	20 to +61 dBmv	23 to 61 dBmv
UPSTREAM 16QAM	20 to +58 dBmv	23 to 56 dBmv

 DOCSIS 3 ERROR STANDARDS

A quick word about “symbols”, the unit into which data is packed.  The higher the bit density, the more efficient the data transport is but the more prone it becomes to errors. The table below shows the density currently used in VM’s cable network.

Modulation	Bits/symbol
QPSK	2
16QAM	4
64QAM	6
256QAM	8

Error correction is a complicated subject including “interleaving” & “codewords”.  Suffice it to say that error correction operates on interleaved codewords; each codeword comprises a number of symbols.

Commonly known as SNR (Signal to Noise Ratio), CMs MUST perform according to the table below by achieving a downstream codeword error rate (CER) ≤ 9 x 10e-7.

Modulation	CM Input Power	SNR	+ 3dB
64QAM	-17 to + 17 dBmv	≥ 25.5 dB	≥ 28.5 dB
256QAM	-13 to -6 dBmv	≥ 34.5 dB	≥ 37.5 dB
256QAM	-6 to +17 dBmv	≥ 31.5 dB	≥ 34.5 dB

The above SNR values are minima laid down by the European DOCSIS 3 spec based on what leaves the CMTS & arrives at a CM.  Headroom must be added of at least 3 dB for impairments en route.

For VM's DOCSIS 3 modems, which are customised and branded to VM, no formal specs are obtainable.  So the DOCSIS 3 spec is the only available guidance against which problem reports can be judged.

.... PART B is in the next post.

[Sephiroth]

POWER LEVELS & SNR: A TECHNICAL PRIMER (Updated November 2011)

PART B - UPSTREAM TRANSMISSION

The so called "fibre-optic broadband" is more accurately referred to as "cable broadband" because at both ends (home & head end), termination is by means of coaxial copper cable.  It is a nominal requirement that each active node (e.g. Optical Node), on the upstream path receives a minimum of 15 dBmv. This 15 dBmv is not rigidly applied because it is obviously impossible for all modems to hit the node at the same power.  There is intermediate upstream amplification.

So let's look at a theoretical (but practically founded) scenario, using the nominal 15 dBmv for illustration. In the diagram, the path to the optical node might comprise the following elements, each contributing to attenuation:

In the example, upstream power would be 48.5 dBmv.  If the coax cable goes to a 10 dB tap point instead of 23 dB, then the upstream power would be 35.5 dBmv, making 15 dBmv at the optical node.  Coax attenuation varies according to upstream frequency (the higher it is, the greater the attenuation but for the small upstream range that isn't a huge problem).  Doubling the coax length doubles the attenuation.

Impairments outside of these "known" attenuations requires extra power to assure 15 dBmv to the optical node.  The CMTS manages this by simply deciding whether or not the CM has been able to communicate properly with the CMTS.  If not, during initial ranging, the CMTS commands a 0.25 dBmv increment until communication is established.  During a session, periodic maintenance opportunities ("keep-alives") occur. If impairments have arisen that disrupt keep-alive communication, the CMTS offers a series of further opportunities, each failure resulting in a T3 message. After 16 such T3 messages, the CMTS removes the modem from its polling list and a T4 timeout occurs with a subsequent modem reset and possible power ramping as per initial ranging.

Modems to a single street cabinet are arranged (through differently attenuated tap points) to have as similar a transmit power as possible so as to preserve SNR; otherwise there would be wide SNR variations between modems close to the street cabinet and those further away.  That said, the CMTS will command upstream power according to the attenuation (and SNR) seen for a given modem.  So you can see why a ramping upstream power number is potentially something to worry about, whereas a stable, but high number without problems is nothing to worry about.

UPSTREAM POWER LEVELS

UPSTREAM
Single channel Upstream	Acceptable	Optimal
VM Recommended	31 to 39 dBmv; 51 to 54 dBmv	40 to 50 dBmv
Seph’s Reasoned Power	31 dBmv to 54 dBmv STABLE	35 to 50 dBmv

The recommended levels are not cast in stone as the technical explanation shows.  Earlier, the power level necessary to reach the first active node was discussed.  The CMTS at the VM end is but a big fat modem and, depending on settings at the local hub node, power entering the line card needs to be similar to the downstream entering a cable modem.  Once the upstream goes past the optical node and along the fibre backbone, it has to be re-modulated to RF so that it enters the line card from a coax connexion.  There should be no noise ingress at the CMTS end – and sometimes there is, which we don’t see.  Upstream noise (SNR) is not reported to the user and is not further dealt with here.

The CMTS commands output power according to the target receive power seen against actual receive power.  Local settings, e.g. cabinet tap point positioning, deal with hitting the first node at the right power level.  If the CM is close to the street cabinet, then you wouldn't expect to see a high upstream power level if your neighbour had a lower level unless you had been placed on a different tap point.  

DOCSIS 3 has additional power demands when driving higher UPSTREAM bandwidth (e.g. 6.4 MHz or bonded channels).  Further reading can be found here.

If the power is fluctuating during initial ranging (including after a T4 event), then the reported 58 dBmv is not enough to establish communication with the CMTS.  The problem could lie on the CMTS side of the network, although most cases are local with a faulty amplifier in the street cabinet or some attenuation on the line to the street cabinet.

THE EFFECTS OF LOW UPSTREAM POWER

This can be tricky.  A house that is close to the street cabinet needs less power than one that is further away, subject to the tap point used at the street cabinet.  This is why upstream values like 29 dBmv can work OK (though there can then be SNR insertion issues) - but it could mean that the house is connected to a low attenuation tap point and thus the downstream power might be too high.

If you are only 20m from the street cabinet instead of 100, you immediately gain 3 dB and if you are on the 10 dB tap, you only need 32.5 dBmv to make 15 dBmv at the optical node. 

So, if you are some distance from the street cabinet AND your upstream power is low (with possible attendant SNR issues) AND you're having problems, then you could have a faulty modem.

.... PART C is in the next post.

[Sephiroth]

POWER LEVELS & SNR: A TECHNICAL PRIMER (November 2011)

PART C - DOWNSTREAM 

In PART B, we examined the upstream leading to a rationale for recommended power values in the various circumstances that pertain.

Here in PART C, we examine the downstream.  It's frequency range is much higher than the upstream and so, is less susceptible to (but not immune from) noise.  The downstream, therefore, works at a much higher modulation (bit packing density) than the upstream.

DOWNSTREAM TRANSMISSION

The CMTS at the VM end puts out upwards of 42 dBmv per channel. The level arriving at the first downstream node is variable depending on plant condition; the output of a particular line card matches those dependencies.  This approach ensures (in a correctly maintained system) that the signal reaches the optical node at 15 dBmv. When the signal emerges at the optical node in the street cabinet, it has to overcome the 33.5 dB built in attenuation (see diagram) on the 23 dB tap point, or 20.5 dB attenuation if connected to the 10 dB tap point.  PLUS another 10 dB due to the much higher downstream frequency.  So that's 43.5 dB or 30.5 dB attenuation for downstream in the two tap point cases.  But 100m further down the road, the coax component of attenuation doubles adding 14 dB (more for higher frequencies), making the example values 57.5 dB and 44.5 dB.

So now you can see why downstream power levels can go negative when the target output of the optical node is 46 dBmv.

THE EFFECTS OF HIGH DOWNSTREAM POWER

High downstream power (say in double figures), occurs if:

• The CM is near to the street cabinet & the illustrated attenuation figures are much lower AND the installation engineer has neither put the cable onto a higher attenuation tap nor added a forward path attenuator at the CM;

• A faulty street amplifier is "overdriving" the power level.

High downstream power amplifies noise (but if the CM is close to the street cabinet, SNR will be high and noise will be low).  High downstream power also overdrives the CM amplifier, but not all CMs are as sensitive as each other.  So this is a moving target.

The DOCSIS 1 modems tune the RF channel they're looking for.  The signal is QAM modulated (which you can read about here). High power levels can cause DOCSIS 1 QAMs around the tuned one to break through the tuner's isolation and interfere with the intended signal.

DOCSIS 3 modems work differently from DOCSIS 1.x modems.  The 1.x modem tunes to an RF channel 6 /8 MHz wide (DOCSIS/EURODOCSIS) whereas the DOCSIS 3 modem digitally samples 100 MHz of the RF.  This additional analogue to digital conversion is reflected in the less high SNRs (RxMER) evident in the DOCSIS 3 modem stats.  There is a good technical read on some of this here.

 DOWNSTREAM POWER & SNR/RxMER LEVELS

Bearing in mind the above explanations, it should be clear why the following levels are recommended.  They take into account

• the modem's specifications
• the DOCSIS 3 specifications
• the technical explanations provided above
• the anecdote on the forums as to where problems arise
• the values based on geography (i.e. network build) recommended by Virgin Media staff

DOWNSTREAM
Modulation & Super Head End	VM Optimal Power	Seph’s Reasoned Optimal Power	VM Optimal SNR	Seph’s Reasoned Acceptable SNR
256QAM	-3 to + 7 dBmv	-3 to + 7 dBmv	≥ 35 dB	≥ 34.5 dB
64QAM Langley SHE	-10 to +0 dBmv	-4 to +5 dBmv	≥ 35 dB	≥ 30 dB
64QAM Knowsley SHE	-7 to +3 dBmv	-4 to +5 dBmv	≥ 35 dB	≥ 30 dB

Looking at the stats on the VM forums, it is clear that power levels below -4 dBmv tend to increase the codeword error count.  For DOCSIS 3, power levels > 7 dBmv have been known to cause synchronisation issues.  However, for DOCSIS 1, power levels on some modems (e.g. Cisco) of 10 dBmv are not problematic.

Also looking at the various stats, it is clear that users on the more noise tolerant 64QAM modulation must have an SNR of at least 30 dB provided that the power levels are within the optimal range.  If the power is high and the SNR at the bottom end, noise is amplified.  I can confirm that my own 50 meg service running currently at 256QAM will not synchronise at 7 dBmv but works well at 34 dB SNR with optimal power levels.

Regarding downstream SNR, don't let anyone tell you that for 256QAM modulation that anything less than 34.5 dB RxMER/SNR is acceptable.  I can publish my recorded VMNG 300 stats - all my RxMER values are between 34 & 36 dB) and show how a small variance of noise in the 34 dB range can increase the rate at which correctable codeword errors arise.  Whilst it is true that correctable errors don't affect your performance, there comes a point where these become uncorrectable which is wholly unacceptable.  By observation on the forums and my own recent experience with both the VMNG300 & the SH, the tipping point for 256QAM at which uncorrectable codeword counts rise is c. 32 dB; this can vary between cable modems and according to home conditions. 

However there are official tables that set out the lower threshold for each QAM level; additionally good engineering practice adds 3 dB for local conditions and best practice adds 6 dB which would be an ideal design goal.  So I'd be wanting to see better than 33 dB at 256QAM and at least 30 dB headroom on 64QAM.  The upstream is most noise sensitive due to its frequency range; hence only 16QAM bit density, reverting to QPSK in poor conditions.  If your stats show QPSK and low downstream SNR, then you usually have a common cause fault, which means that a component that handles both your upstream and downstream (e.g. coax cable, cable connector, modem) is electrically noisy; you'll have high correctable codewords (VMNG 300) shown and your web pages will take some time to load.

Note that for the upstream a SNR threshold is set in the CMTS for each cable, below which modulation is dynamically set to QPSK.  You are unaware of the upstream SNR for your circuit.

SNR/RxMER Thresholds

Virgin Media show a lot of concern for the upstream and set their target SNR for received by the CMTS at +6dB headroom.

Modulation	Lower SNR Threshold	+ 3dB Headroom	+6 dB Headroom
QPSK	07 to 10 dB	13 dB	18 dB
16QAM	15 to 18 dB	21 dB	24 dB
64QAM	22 to 24 dB	27 dB
256QAM	28 to 30 dB	33 dB

END NOTE

The Virgin cable network is a collection of different network builds by different companies originating in the 1990s. These builds vary in quality and hence settings at the CMTS.  Some are on DOCSIS (5 - 42 MHz upstream); others (later) on EURODOCSIS (5 - 65 MHz, soon going up to 85 MHz).  The upstream frequencies are a noisy part of the spectrum, particularly the lower frequencies which thus tend to be QPSK modulated. If plant is improved, higher capacity line cards are installed, software based re-segmentation is introduced, then better than 16QAM can be achieved with consequently higher data rates. 

For the downstream, the same changes as for upstream and expanding the upper frequency range, will yield additional bonding channels to support 200 meg downstream.

END.