IEEE 802.14-96/015

TITLE: Non-Linear INTRA for Upstream CATV Modems
AUTHOR: Bill Miller

ABSTRACT: After an analysis of the generic modulation types currently under study for Upstream CATV, some preliminary simulation results for a simplified FM design running at only 10.5 Mb/s are presented. The following paper is based on Document # IEEE 802.14-96/015 prepared to assist and submitted to the IEEE 802.14 WG Cable TV Protocol Working Group January 14, 1996.

Generic Modulations

Two generic modulation types have been proposed for upstream use in the 802.14 WG----INTRA and n-QAM (n=4 or 16). After an analysis of the generic modulation types currently under study for Upstream CATV, preliminary simulation results for a simple FM Modem operating at 10.5 MHz are presented.

The PHY layer should provide:

Low cost to the set-top box owner
High revenue potential to the cable system owner
Performance that justifies the subscriber's cost
The PHY layer's contribution to each of these three goals seems to come from two main factors:

Table 1. The PHY layer affects some reasonable goals
GOAL	FACTOR A	FACTOR B
Low Set-top Box Cost	Number of RF Components	Digital Complexity of PHY
High Operator Revenue	Modulation Efficiency	RFI Noise Strategy
High Performance PHY	Layer Data Rate & BER	PHY Layer Overhead

The revenue from maximum utilization of the limited upstream bandwidth resource is assumed to be more important to the cable system operator than the cost of the Headend equipment. Table 2. summarizes the proposals made at the Nov. 1995 meeting. Formula 1 is in 95/094, where it is shown that FM-SSB can have a better Bandwidth Efficiency ("formula 1") than the BW Efficiency of n-QAM ("formula 2") at equivalent receiver SNR. Multiple narrow-channel QPSK RF modems are indicated in Table 2, assuming that the MAC layer for a 21st century system needs perhaps 10 Mb/s to provide commercially competitive performance to 500 to 2000 subscribers (see Functional Requirements). The wide-channel modems require only one RF circuit. The RFI strategy of CDMA and DMT-type systems is to "eliminate" the code-vectors or tones that are experiencing RFI, which effectively abandons a portion of the precious upstream resource in opposition to GOAL 2. Since it uses sub-bands (coordinates), the baseband INTRA of the FM Modem can employ the elimination strategy, if necessary. But in addition, passband FM should also have some immunity to RFI. In principle, all the proposals could avoid regions of the spectrum, or degrade the performance (bit rate or BER) in severe interference. These strategies affect the PHY Layer overhead.

Table 2. Comparison of modulation proposals for upstream CATV
Proposal Class	Modulation Type	Channel Bandwidth	Number of RF Circuits	Digital Complexity	Modulation Efficiency	RFI Noise Strategy
INTRA	FM	Wide	Low	Low	Formula 1	Immunity
CDMA	n-QAM	Wide	Low	Low	Formula 2	Elimination
DMT	n-QAM	Wide	Low	High	Formula 2	Elimination
7 Others	n-QAM	Narrow	High	Low	Formula 2	Avoidance

A Simple FM Modem

The robustness of Linear INTRA to harmonically related noise was demonstrated in 802.14-96/014. The Non-Linear INTRA modem combines the Linear INTRA baseband modulation with an FM passband modulation. Some representative parameters for FM modems were given in 802.14-95/094, where it was shown that the modulation index can be optimized to provide the highest data rate for a given CNR, depending on the gain obtained from non-linear companding and de-emphasis in the receiver. The calculations in 95/094 show that INTRA FM-DSB can have a significantly higher bandwidth efficiency than n-QAM at equivalent receiver input SNR. An INTRA FM modem without companding was simulated as in Figure 1.

 

Figure 1. A simplified simulations block diagram for INTRA FM

The LOG-AMP amps, equalizers, etc. associated with companding could be inserted at X and Y for a CATV system, but a simple modem can achieve useful data rates without them. Also it is a design choice as to whether the FM modulator and demodulator actually operate at the passband, or at an intermediate frequency. Which blocks on the right are digital and which are analog is a further design choice affecting the cost of analog converters etc. An all digital set-top box upstream modulator appears to be possible with off-the shelf components.

Simulations

Uniformly random data was generated and partitioned into 200 vectors carrying 4 data coordinates. The data coordinates of the vectors were then digitally pre-emphasized as described in 802.14-95/094. However, the 4-Dimensional modem transmitter design was simulated with an 8-Dimensional rotation, which serves as a x2 Upsampler. This keeps the baseband data away from the interpolation filters of the simulator's "RESAMPLE" routine so no equalization filter (and no adaptive rotation) were required. The receiver's 8-D rotation provides a x2 Downsampling so the upper half of the parabolic noise spectrum in the receiver is visible to observe the demodulated RFI from the Channel Noise Model. The bit rate is 10.5 Mb/s. The FM modulation index was chosen as 1, which is not the optimum value for highest data rate. The pre-emphasized INTRA signal (point X of Fig 1) is shown in Figure 2.

Figure 2. Pre-emphasized INTRA at 10.5 Mb/s (m=1)

The N90ile noise model was used and the input to the Discriminator is shown in Figure 3.

Figure 3. INTRA FM-DSB without companding (m=1)

The discriminator output (point Y of Fig 1) is shown in Figure 4.

Figure 4. Receiver's Baseband at 10.5 Mb/s

The receiver's rotation produced the following for the standard deviation from threshold:

 

Table 3. INTRA FM-DSB with non-optimized modulation index (m=1)
Data Coordinate	1	2	3	4
Pre-emphasis	x1	x2	x4	x8
Data Bits	5	4	3	2
Std Dev T-7	5.3 %	7.3 %	6.2 %	4.1 %
Std Dev T-8	8.4 %	10.2 %	8.7 %	5.7 %
Std Dev T-9	4.7 %	5.6 %	4.5 %	2.8 %
Std Dev T-10	4.1 %	5.0 %	3.4 %	2.5 %
Std Dev T-11	3.6 %	5.0 %	3. 5 %	2.4 %
Std Dev T-12	4.0 %	4.9 %	4.0 %	2.6 %
Std Dev T-13	3.2 %	4.7 %	3.6 %	2.3 %

Table 3a . CATV channels (USA standard)
Channel	T-7	T-8	T-9	T-10	T-11	T-12	T-13
Center (MHz)	8.75	14.75	20.75	26.75	32.75	38.75	44.75

The standard deviation of the error margin (see 802.14-96/014) gives a BER >> 10^-8. The simulation did not use a Limiter before the discriminator. Note that the equations for the pre-emphasis gain include the factor (Hi^2)/(3I^2+3I+1). For the convenient binary amplifications 1,2,4,8 shown in the table, this factor evaluates to 0 DB, -2.4 DB, -0.75 DB and +2.4 DB. Thus coordinate 4 will have much less error than coordinate 2 even where there is no RFI. Also there is "quantization noise" because the INTRA signal (at point X and Y in Figure 1) has been truncated to 11 bits as shown in Figure 5. With the noise model disabled, the error margins per coordinate are 2.3 %, 2.9 %, 1.3 % and 1.0 %, so a 12-bit digital modulator (or D/A) could be used in the set-top box.

Figure 5. The INTRA baseband signals (X & Y) used 11-Bits resolution