While North America developed B8ZS for its T-carrier systems, the rest of the world took a different path. The CCITT (now ITU-T) developed HDB3 (High-Density Bipolar 3-zeros) as the international standard for zero suppression in E-carrier systems and beyond.

HDB3 embodies the same fundamental insight as B8ZS—using intentional bipolar violations to signal zero sequences—but with a crucial difference: HDB3 substitutes after only 4 consecutive zeros, compared to B8ZS's 8. This more aggressive approach provides stronger synchronization guarantees but requires more sophisticated encoding logic.

Understanding HDB3 is essential for anyone working with international telecommunications, as it is the standard used in E1 (2.048 Mbps), E2 (8.448 Mbps), E3 (34.368 Mbps), and higher-order systems throughout Europe, Asia, Africa, and South America.

HDB3 builds on the same AMI (Alternate Mark Inversion) foundation as B8ZS, but with different design choices:

Key Differences from B8ZS:

The HDB3 Design Philosophy:

HDB3 targets a maximum of 3 consecutive zeros (hence the '3' in HDB3). By acting at 4 zeros rather than 8, HDB3 provides:

1. Stronger clock recovery — Maximum zero run is 3 instead of 7 (B8ZS allows runs of 3 zeros within its substitution)
2. More frequent transitions — Guaranteed pulse density is higher
3. Better suited for higher bit rates — Critical when bit periods are shorter

The tradeoff is increased complexity and slightly more frequent substitution overhead.

Understanding the Substitution Logic:

HDB3's substitution patterns are more nuanced than B8ZS because HDB3 must maintain DC balance while using only 4 bit positions. The key insight is that HDB3 uses two types of substitution patterns:

Pattern Type 1: 000V (when ones count is odd)

• Three zeros followed by a violation pulse (V)
• The violation has the same polarity as the preceding pulse
• Only one non-zero symbol in 4 positions
• Does NOT maintain DC balance on its own

Pattern Type 2: B00V (when ones count is even)

• A balance pulse (B) followed by two zeros and a violation
• B has the opposite polarity of the preceding pulse (following AMI rule)
• V has the same polarity as B (which is a violation)
• Two non-zero symbols that are the same polarity
• The B and V cancel out, maintaining DC balance

Why Two Patterns?

The clever design ensures overall DC balance:

• After a 000V substitution, one pulse was added
• If ones count was odd before, it becomes even, and vice versa
• This alternates which pattern type is used
• Over time, B00V and 000V patterns alternate, and the B pulses maintain balance

HDB3 encoding requires tracking two state variables:

1. Last pulse polarity: + or - (the polarity of the most recently transmitted non-zero symbol)
2. Ones count since last substitution: Even or odd (counting data 1s, not including B pulses from substitutions)

The Encoding Algorithm:

Initialize:
  last_polarity = undefined (first pulse can be either)
  ones_count = 0

For each input bit:
  If bit = 1:
    Transmit pulse with polarity opposite to last_polarity
    Update last_polarity
    ones_count = ones_count + 1
  
  If bit = 0:
    Add to zero_run_buffer
    
    If zero_run_buffer has 4 zeros:
      If ones_count is odd:
        Transmit 000V (V has same polarity as last_polarity)
        Update last_polarity to V's polarity
      Else (ones_count is even):
        Transmit B00V (B is opposite of last_polarity, V is same as B)
        Update last_polarity to V's polarity
      
      ones_count = 0
      Clear zero_run_buffer
    Else:
      (Wait for more bits)

Detailed Example:

Let's encode: 1 0 0 0 0 1 0 0 0 0 0 0 0 0 1

This contains two groups: first 4 zeros, then 8 zeros (which will trigger two substitutions).

Step 1: Bit 1 (data 1)

• First pulse, choose + (either is valid)
• Transmit: +
• State: last_polarity = +, ones_count = 1

Step 2-5: Bits 2-5 (0000)

• Zero counter reaches 4
• ones_count = 1 (odd) → Use 000V pattern
• V has same polarity as last_polarity (+)
• Transmit: 0 0 0 +
• State: last_polarity = +, ones_count = 0 (reset)

Step 6: Bit 6 (data 1)

• Data 1, pulse should be opposite of last_polarity
• Transmit: -
• State: last_polarity = -, ones_count = 1

Step 7-10: Bits 7-10 (0000)

• Zero counter reaches 4
• ones_count = 1 (odd) → Use 000V pattern
• V has same polarity as last_polarity (-)
• Transmit: 0 0 0 -
• State: last_polarity = -, ones_count = 0

Step 11-14: Bits 11-14 (0000)

• Zero counter reaches 4 again
• ones_count = 0 (even) → Use B00V pattern
• B is opposite of last_polarity: +
• V is same as B: +
• Transmit: + 0 0 +
• State: last_polarity = +, ones_count = 0

Step 15: Bit 15 (data 1)

• Data 1, pulse should be opposite of last_polarity
• Transmit: -
• State: last_polarity = -, ones_count = 1

HDB3 decoding exploits the fact that violations are recognizable:

Decoding Algorithm:

For each received symbol:
  If symbol is 0:
    Buffer the zero
    
  If symbol is non-zero (+ or -):
    Check if this is a bipolar violation (same as last pulse)
    
    If violation detected:
      This is a V pulse
      Look back for B00V or 000V pattern:
        If 3 positions back is non-zero: B00V pattern
          Replace last 4 symbols with 0000
        Else: 000V pattern
          Replace last 4 symbols with 0000
    Else:
      Normal pulse = output 1
      Output any buffered zeros (up to 2)
    
    Update last polarity tracking

Decoding the Previous Example:

Received: + 0 0 0 + - 0 0 0 - + 0 0 + -

Final decoded output: 1 0 0 0 0 1 0 0 0 0 0 0 0 0 1

✓ Matches original data exactly!

Distinguishing B00V from 000V:

The decoder determines which pattern was used by examining the symbol 3 positions before the violation:

• If zero: 000V pattern was used
• If non-zero: B00V pattern was used

This works because:

• In 000V, positions 1-3 are always zeros
• In B00V, position 1 is always a pulse (the B pulse)

Handling Ambiguity:

Could a normal AMI sequence look like a B pulse? Consider a data sequence 1 0 0 0 0:

• The leading 1 generates a pulse (say +)
• The four zeros trigger substitution
• If ones_count is even, B00V is used: B (which would be -) 0 0 V (which would be -)

But wait—the incoming data '1' just set ones_count to 1 (odd), so 000V would be used instead!

The algorithm ensures that whenever a real data '1' occurs immediately before zeros, ones_count is odd, and 000V is used. B00V only occurs when the preceding positions were also substitution or the count accumulated to even.

HDB3's DC balance mechanism is more subtle than B8ZS's and deserves detailed analysis.

The Challenge:

With only 4 bit positions, HDB3 can't guarantee internal DC balance within each substitution like B8ZS does with its 2+2 pattern. Instead, HDB3 achieves DC balance over multiple substitutions.

Analyzing 000V:

The 000V pattern contains one pulse:

• If V is +, contribution to DC is +1
• If V is -, contribution to DC is -1

Without correction, consecutive 000V patterns could accumulate DC:

000+ 000+ 000+ → DC accumulation of +3

The B Pulse's Role:

The B00V pattern contains two pulses of the same polarity:

• If B is + and V is +, contribution is +2
• If B is - and V is -, contribution is -2

This seems worse! But the key is when B00V is used.

The Balancing Act:

Consider what happens with long zero runs:

With continuous zeros, B00V alternates polarity, and DC oscillates between +2 and 0, never accumulating unbounded.

With Mixed Data:

When data contains ones between zero runs, the pattern changes:

Why It Works:

The mathematical insight is:

1. Data 1s alternate polarity by AMI rules, contributing net zero DC over time

2. 000V patterns inherit the last polarity, so they track the alternation

3. B00V patterns occur when ones_count is even, meaning an equal number of + and - data pulses have occurred since the last substitution

4. B pulses follow AMI rules, so they also alternate

5. V pulses in B00V match B, creating pairs that alternate each substitution

Proof of Bounded DC:

Let's prove that DC never exceeds bounded limits:

• After any 000V: DC has shifted by ±1
• After any B00V: DC has shifted by ±2
• But B00V alternates in polarity (because last_polarity flips with each substitution)
• And 000V is only used when there's an odd number of data 1s, which inherently means the alternation continues

Maximum DC deviation is bounded by 2 pulse units, regardless of input data.

Understanding the differences between HDB3 and B8ZS illuminates the design tradeoffs in zero suppression techniques.

Historical Context:

The different approaches reflect different evolutionary paths:

T-Carrier (North American) Evolution:

• Started with strict ones density requirements
• Gradually relaxed with B8ZS
• 8-zero trigger provided compatibility buffer
• Infrastructure already in place when B8ZS deployed

E-Carrier (International) Evolution:

• Developed slightly later with more aggressive design
• HDB3 specified from early stages
• 4-zero trigger reflected higher reliability requirements
• Clean-sheet design allowed more complex algorithm

Why Both Survive:

Neither standard displaced the other because:

1. Installed base — Billions of dollars in infrastructure
2. Regional standards — Different regulatory bodies
3. Interworking — Gateways convert between systems
4. Functional equivalence — Both achieve the goal adequately

Modern equipment often supports both, configurable for the region.

Understanding how HDB3 operates within E1 systems provides practical context for its application.

E1 System Overview:

The E1 carrier system transmits 2.048 Mbps (slightly faster than T1's 1.544 Mbps):

• 32 time slots × 8 bits per slot × 8,000 frames/second = 2.048 Mbps
• Slot 0: Frame alignment and signaling
• Slot 16: Signaling (if CAS signaling used)
• Slots 1-15, 17-31: 30 traffic channels

Unlike T1's 24 channels, E1 provides 30 user channels—a 25% capacity advantage.

HDB3 Application Points:

Traffic Data → E1 Framing → HDB3 Encoding → AMI Line Transmission

HDB3 operates at the physical layer, processing the complete 2.048 Mbps bit stream including frame overhead.

Frame Alignment Word (FAW):

E1 uses a Frame Alignment Word in time slot 0 for synchronization:

• Frame 0: 0011011 (FAW)
• Frame 1: 1xxxxxx (Not-FAW, with alarm bits)

The FAW pattern is carefully chosen:

• Contains transitions
• Unlikely to appear in random data
• Works well with HDB3 encoding

CRC-4 Multiframe:

Modern E1 uses CRC-4 for error monitoring:

• 16 frames grouped into a multiframe
• CRC calculated over submultiframe
• Provides error detection for monitoring

HDB3 is applied independently of the CRC-4 structure—it operates on the continuous bit stream regardless of higher-level framing.

While E1 is HDB3's most common application, the technique is used throughout the PDH (Plesiochronous Digital Hierarchy) at multiple rates.

*CMI = Coded Mark Inversion, a different technique used at E4 and above

Why HDB3 at E2 and E3?

At higher bit rates (E2: 8.448 Mbps, E3: 34.368 Mbps), the challenge of clock recovery intensifies:

• Shorter bit periods require faster clock recovery
• Higher frequencies are more susceptible to jitter
• Longer cables accumulate more timing degradation

HDB3's guarantee of maximum 3 consecutive zeros remains critical at these rates. The algorithm scales directly—the same encoding rules apply, just at faster rates.

E3 Specifically:

E3 (34.368 Mbps) carries 16 E1s or 480 voice channels:

Bit period: 29.1 ns
Maximum zero run (with HDB3): 3 bits = 87.3 ns
Without zero suppression: Unlimited → clock loss

At this speed, clock recovery circuits are already challenged. Without HDB3, even moderate zero runs would cause problems.

The Transition to CMI:

At E4 (139.264 Mbps) and above, HDB3 is replaced by CMI (Coded Mark Inversion):

• CMI encodes 1s as alternating full-bit high/low
• CMI encodes 0s as stable half-bit low, half-bit high
• Every bit period contains at least one transition
• No substitution needed—clock is inherent

CMI's advantage at very high speeds is its guaranteed transition every bit period, eliminating any dependence on data patterns.

We have thoroughly explored HDB3, the international standard for zero suppression. Let's consolidate the essential knowledge:

Looking Ahead:

With B8ZS and HDB3 mastered, you understand the two major substitution-based zero suppression techniques. The next page explores Randomization—the mathematical foundations of pseudo-random sequence generation that underpin more general scrambling techniques used in modern high-speed systems.

Randomization extends beyond simple substitution to transform entire data streams, enabling even stronger guarantees for clock recovery and spectral properties.